03 Reliability of Generation Systems

3 Reliability of Generation Systems
The generated electric energy, which is transferred and distributed to the

customers, should be of good quality (taking the voltage and frequency into
account) and have a sufficient reliability level. Operating availabilities of
the generation system plays an important role in meeting these demands. In
past years, a high level of power system reliability has been assured by
careful planning and operating practices based on experience and
engineering judgment. The results were satisfactory as long as power
systems were relatively small and simple. With the increasing size and
diversity of power systems, the problems became much more complicated.
This necessitated developing new methods for determination of planning
and operation criteria, based on more sophisticated techniques and the
reliability theory.
Using detailed mathematical models and the probabilistic approach, new
methods have been developed in the last decade. They permit assessing the
generation system reserve which is required to provide a predetermined
reliability level.
The reliability indices should be determined for the entire system, i.e.,
the generation system, the transmission and the distribution networks. The
complicated nature of a power system makes this problem very difficult.
Consequently, most of the reliability studies have dealt with only one
subsystem, assuming that there are no reliability problems with the rest of
the system. In actual practice, this assumption is especially true for the
generation part of the entire system. In this case the required generating
capacity reserve for a given load can be determined together with techno-
economical aspects (Allan and Billinton 1992). Today the planner has
sufficient tools to evaluate the generation system reliability in terms of
probabilistic reliability indices, such as the loss of load probability.
The system indices have been similarly assessed through the transmis-
sion network However, such solutions are not accurate enough to be used
for the design of the transmission network (unlike the generation system).
This is due to the strong dependency between the generation and trans-
mission systems. Therefore, the correct approach is to study the composite
reliability of both systems.
In the current chapter, the reliability indices of the generation system are
studied. With them the required generation reserve can be determined
under the assumption that the other parts of the power system are fully
reliable. The reliability of the generation system can be characterized by
the existing redundancy level, which is influenced by several factors such
D. Elmakias: Reliability of Generation Systems, Studies in Computational Intelligence (SCI) 111,

171–238 (2008)
www.springerlink.com © Springer-Verlag Berlin Heidelberg 2008
172 New Computational Methods in Power System Reliability
as, for example: the required load level or the reliability of the individual
generating units. Clearly, the higher the reliability level of the individual
units, the higher will be the reliability level of the whole generation system.
Generally speaking, the reliability level of any system is determined by
the risk that under certain conditions the system will not operate
satisfactorily due to failures in the system elements. In the generation
system these indices can be determined by a relatively simple approach,
namely by comparison of the required load to the generation capacity in the
probabilistic manner.
As in an entire system, its generation part reliability is described with the
aid of two types of such risk indices. The first type includes the risk
relating to the static condition of the system, i.e., the system behavior at
steady state. The second type includes the risk for the dynamic conditions,
i.e., the system behavior under transient phenomena. Thus, the static
indices do not relate to the transient phenomenon starting at the failure
occurrence and ending at the new steady state condition. If, for example,
the required reserve is studied, one should differentiate between operational
reserve (which includes the spinning or hot reserve and a quick start of gas
turbines) and the installed reserve capacity. The operational reserve relates
to the dynamic condition of the system, i.e., to the system operation, while
the installed reserve relates to the system design which determines the
number and size of the units and the time of their operation in response to
increasing electricity demand.
The static and dynamic system conditions can be illustrated by the
following example. Assume that one of the large generating units is out of
operation due to a failure, when the load level is, say, at peak. At that
moment, the system enters a transient condition, where the controls of
frequency react immediately. If the frequency rate drops or the frequency
itself reaches certain predetermined values, a load shedding procedure (i.e.,
disconnection of certain consumers) takes place. Such a procedure may
take a few seconds. At that point, the dispatcher intervenes by, for example,
“throwing in” of gas turbines, which takes minutes to start operating. Later,
if required, hot units can be committed in the system. This may take hours.
Finally, the power system will reach a new steady state, where there may
still be unsupplied consumers due to shortage in generating capacity.
Two distinct states characterize the above procedure, namely:
1. The dynamic state, which starts at the moment of failure and
continues throughout the procedure of adding generating units,
2. The static state, which is the new steady state.
The reliability indices for the generation system at steady state are
discussed first. The installed reserve, which is required to provide the
Reliability of Generation Systems 173
system operation at a predetermined risk level, can be studied by means of

the abovementioned indices. In this case it is commonly accepted to define
the installed reserve as the difference between the total installed capacity of
all generating units and the annual peak load.
This is an important tool for the planner for determination of the number
of generating units, their size and commission dates into the system in
order to meet the expected load demand.
The risk level can be calculated in a similar way even for the dynamic
state, when a part of the required load may be non-supplied during the time
interval between the failure occurrence and the new steady state. The
results of this study are of great importance to the system operator in order
to define a priori the immediate operational reserve required for a known
daily load at a predetermined risk level. The design of the generation
system in the past was based on deterministic reliability criteria. In this
approach, the reserve margin (RM) is defined as the difference between the
total available generation capacity GA, and the annual peak load Lp:
GA − Lp
RM= 100% . (3.1)
Lp
This index is not probabilistic. It has been used until new models based
on probabilistic reliability indices were developed. As for the reserve, it
was determined by accumulated praxis. It should be noted that the RM
index does not take into account the type of the generating units (thermal,
gas turbines, etc.), their sizes and their reliability levels. Thus, two systems
with markedly different reliability levels may have the same RM value. As
an example, consider two systems with the same annual peak load value
and the same RM. The first system is composed of large generating units
and the second of smaller units, say 50% of the large units. It is obvious
that the second system is characterized by higher reliability level. The
reason for that is that the probability for one large unit outage is higher than
the probability for a simultaneous outage of two small units.
Another disadvantage of the deterministic indices is that they ignore the
stochastic nature of factors influencing the reliability level, like, for
example, the available generation capacity or the predicted load level. As a
result, the deterministic indices overestimate the system reliability in
contrast to a more sound probabilistic approach.
Accurate reliability evaluation should take into account the unit
commitment and given operational constraints. The combination of a
thorough mathematical analysis with computer-based computations gives
not only more realistic indices values by simulating the unit commitment
through randomly generated availabilities of the units but also allows for
working out the cost-saving reserve management during system operation.

The enhanced features of the modern approach are detailed below.
3.1 Reliability indices of generation systems
The calculation of reliability indices for the steady state, required for the
planning of the generation system is based nowadays on two analytical
methods. These methods enable to determine the required reserve for the
generation system, taking into account the predicted load behavior and a
predetermined reliability level. The first method is the loss of load proba-
bility, which means the probability of the load being larger than the
available generation. The second method enables, besides the determi-
nation of the probability, to calculate the frequency and duration where the
load is larger than the available generation. In this method, the frequency
and duration, besides the probability, are studied. It is called the Frequency
and Duration Method. The main difference between these two methods lies
in the load models and the reliability parameters of generating units.
Both methods mentioned above are widely used, though the first one is
more “popular”. It should be mentioned that there is a third method for
calculation of reliability indices, based on Monte Carlo simulation. All
these methods are presented and discussed in this chapter.
3.1.1 Calculating reliability indices using the loss of load

probability method
In order to calculate the reliability indices of the generation system, it is

assumed that the rest of the power system (transformation, transmission,
distribution, etc.) is of the highest reliability level. This means that any
unavailability of power supply to the consumers is due to the generation
system only. In other words, there are no constraints in transferring electric
power via the transmission and transformation systems to the load for any
given condition of the generation system. Such constraints may be, for
example, overloading of transmission lines, voltage drop problems, etc.
The problem under discussion is thus illustrated in Fig. 3.1
In the steady state operation of the system there should be a balance
between the total operating generation capacity G and the load L: G = L. In
the case of a failure in the generation system, the load may exceed the
available generation: G<L. The failure condition (relation between the
generation and the load corresponding to the failure states) can be

represented by a load loss function LLF:
⎧1, if G < L
LLF(G, L) = 1(G < L) = ⎨ (3.2)
⎩0, if G ≥ L.
G
G1 ~
G2 ~ L
:
Gn ~
Fig. 3.1. The simple model of the generation system and the load (G1,..., Gn are
nominal generating capacities of n generating units, L is the load at a certain time)
The LLF produces a binary value for any combination of G and L. The
expected value of the random binary LLF is equal to the probability that
LLF=1 or, in other words, to the probability of the generation system
inability to supply the load. This index is usually referred to as LOLP (Loss
of Load Probability):
LOLP(G, L) = E{LLF(G, L)} = Pr{LLF(G, L) = 1} = Pr{G < L} . (3.3)
The total available generation capacity G usually takes on discrete values
and, therefore can be represented by a pmf gj, pj=Pr{G=gj} for 1≤j≤n. The
discretized load curve is also represented by its pmf li, qi=Pr{L=li} for
1≤i≤k. Having the pmf of the two random variables one can obtain the pmf
of their function LLF(G,L). For this purpose the pmf of G and L should be
represented by u-functions
n gj
U G ( z) = ∑ p jz (3.4)
j =1
and
k
U L ( z) = ∑ qi z l i . (3.5)
i =1
Applying the composition operator ⊗ over UG(z) and UL(z) one obtains
LLF
the u-function representing the pmf of LLF(G,L):

n gj k
U LLF ( z ) = U G ( z ) ⊗ U L ( z ) = ( ∑ p j z ) ⊗ ( ∑ qi z li )
LLF j =1 LLF i =1
(3.6)
n k 1( g j ,li )
= ∑ ∑ p j qi z .
j =1i =1
The expected LLF(G,L) or the LOLP can now be obtained as

n k
LOLP = E (LLF) = U ' LLF (1) = ∑ ∑ p j q i LLF( g j , l i ). (3.7)
j =1i =1
The Loss of Load Expectation (LOLE) represents the expected time (in
hours or days) where the load is larger than the available generation.
Assume that the operation time T is divided into S equal intervals δt=T/S
(time units). In each interval s the load takes on a random value of L (it is
assumed that during the interval the load does not vary). The random time
when the constant load L is not supplied in the interval s is equal to
τs=δt⋅LLF(G,L). This variable can take on either value of δt or 0. The
random time when the variable load is not supplied during the entire
operation period T can be obtained as
S S S T T S
∑τ s = ∑ δt ⋅ LLF (G, L) = ∑ ⋅ LLF (G, L) = ∑ LLF (G, L). (3.8)
s =1 s =1 s =1 S S s =1
The expectation of this time is

⎛ S T ⎞ T
LOLE = E ⎜⎜ ∑ ⋅ LLF (G, L) ⎟⎟ = ⋅ S ⋅ E ( LLF (G, L)) = T ⋅ LOLP. (3.9)
⎝ s =1 S ⎠ S
Another important index named Expected Load Not Supplied (ELNS)
can be obtained by introducing the following unsupplied load function
UL(G,L)=max{L-G,0}. (3.10)
The u-function representing the pmf of UL(G,L) can be obtained using
the ⊗ operator over UG(z) and UL(z)
UL
n gj k
UUL ( z ) = U G ( z ) ⊗ U L ( z ) = ( ∑ p j z ) ⊗ ( ∑ qi z li )
UL j =1 UL i =1
(3.11)
n k max{li − g j ,0}
= ∑ ∑ p j qi z .
j =1i =1
The Expected Unsupplied Load (EUL) is the expected value of the

function UL(G,L). It can be obtained as
n k
EUL = E ( UL) = U 'UL (1) = ∑ ∑ p j qi UL( g j , li )
j =1i =1
(3.12)
n k
= ∑ ∑ p j qi max{l j − gi ,0}.
j =1i =1
The Expected Unsupplied Energy (EUE) is obtained as the product of

the EUL and the loss of load expectation:
EUE = EUL ⋅LOLE. (3.13)
Example 3.1
Consider two separate generating systems (power plants) with a nominal
capacity of 100 MW. In the first system, some types of failure require its
capacity G1 to be reduced to 60 MW and other types lead to a complete
outage. In the second generator, some types of failure require its capacity
G2 to be reduced to 80 MW, others lead to a capacity reduction to 40 MW,
and the rest lead to a complete outage. The generators are repairable and
each of their states has a steady-state probability.
Both generators should supply a variable two-level load L. The high
level (day) load is 50 MW and has the probability 0.6; the low level (night)
load is 30 MW and has the probability 0.4.
The capacity and the load can be presented as a fraction of the nominal
generator capacity. There are three possible relative capacity levels that
characterize the performance of the first generating system:
g10 = 0.0, g11 = 60/100 = 0.6, g12 = 100/100 = 1.0, and four relative
capacity levels that characterize the performance of the second generating
system:
g20 = 0.0, g21 = 40/100 = 0.4, g22 = 80/100 = 0.8, g23 = 100/100 = 1.0
Assume that the corresponding steady-state probabilities are
p10 = 0.1, p11 = 0.6, p12 = 0.3
for the first generating system and
p20 = 0.05, p21 = 0.35, p22 = 0.3, p23 = 0.3
for the second generating system and that the load distribution is
l1 = 50/100 = 0.5, l2 = 30/100 = 0.3, q1 = 0.6, q2 = 0.4.
The probabilistic distributions of the generation capacities and the load
are presented in Table 3.1.
Table 3.1. Distributions of the generation capacities and the load

State 1 State 2 State 3 State 4
First generating g 1.0 0.6 0 -
system p 0.3 0.6 0.1 -
Second generating g 1.0 0.8 0.4 0.0
system p 0.3 0.3 0.35 0.05
Load l 0.5 0.3 - -
q 0.6 0.4 - -
The u-functions representing the capacity distribution of the generating

systems (the pmf of random variables G1 and G2) take the form
UG1(z) = 0.1z0+0.6z0.6+0.3z1, UG2(z) = 0.05z0+0.35z0.4+0.3z0.8+0.3z1
and the u-function representing the load distribution takes the form
UL(z) = 0.6z0.5+0.4z0.3
The expected steady-state capacity of the generating systems can be
obtained directly from these u-functions:
E (G1 ) = U 'G1 (1) = 0.1 × 0 + 0.6 × 0.6 + 0.3 × 1.0 = 0.66
which means 66% of the nominal generation capacity for the first
generating system, and
E (G2 ) = U 'G 2 (1) = 0.05 × 0 + 0.35 × 0.4 + 0.3 × 0.8 + 0.3 × 1.0 = 0.68
which means 68% of the nominal generation capacity for the second
generating system.
The u-functions corresponding to the pmf of the LLF are obtained using
the composition operator:
U LLF 1( z ) = U G1( z ) ⊗ U L ( z )
LLF
0 0.6
= (0.1z + 0.6 z + 0.3z1) ⊗ (0.6 z 0.5 + 0.4 z 0.3 )
LLF
= 0.06 z1(0<0.5) + 0.36 z1(0.6< 0.5) + 0.18 z1(1< 0.5) + 0.04 z1(0< 0.3)
+ 0.24 z1(0.6< 0.3) + 0.12 z1(1< 0.3) = 0.06 z1 + 0.36 z 0 + 0.18 z 0 + 0.04 z1
+ 0.24 z 0 + 0.12 z 0 = 0.9 z 0 + 0.1z1
U LLF 2 ( z ) = U G 2 ( z ) ⊗ U L ( z )
LLF
0 0.4
= (0.05 z + 0.35 z + 0.3z 0.8 + 0.3z1) ⊗ (0.6 z 0.5 + 0.4 z 0.3 )
LLF
= 0.03z1 + 0.21z1 + 0.18 z 0 + 0.18 z 0 + 0.02 z1 + 0.14 z 0

+ 0.12 z 0 + 0.12 z 0 = 0.74 z 0 + 0.26 z1
The system LOLP (expected LLF) is
LOLP1 = U 'LLF1 (1) = 0.1
LOLP2 = U 'LLF 2 (1) = 0.26
The u-functions corresponding to the pmf of the UL(G,L) function are

obtained using the composition operator:
UUL1 ( z ) = U G1 ( z ) ⊗ U L ( z )
UL
= (0.1z 0 + 0.6 z 0.6 + 0.3z1 ) ⊗ (0.6 z 0.5 + 0.4 z 0.3 )

UL
= 0.06 z max(0.5 − 0,0) + 0.36 z max(0.5 − 0.6,0) + 0.18 z max(0.5 −1,0)

+ 0.04 z max(0.3 − 0,0) + 0.24 z max(0.3 − 0.6,0) + 0.12 z max(0.3 −1,0)
= 0.06 z 0.5 + 0.36 z 0 + 0.18 z 0 + 0.04 z 0.3 + 0.24 z 0 + 0.12 z 0
= 0.06 z 0.5 + 0.04 z 0.3 + 0.9 z 0
UUL 2 ( z ) = U G 2 ( z ) ⊗ U L ( z )
UL
= (0.05 z 0 + 0.35 z 0.4 + 0.3 z 0.8 + 0.3 z1 ) ⊗ (0.6 z 0.5 + 0.4 z 0.3 )
UL
= 0.03z 0.5 + 0.21z 0.1 + 0.18 z 0 + 0.18 z 0 + 0.02 z 0.3 + 0.14 z 0

+ 0.12 z 0 + 0.12 z 0 = 0.03z 0.5 + 0.21z 0.1 + 0.02 z 0.3 + 0.74 z 0
The Expected Unsupplied Load is

EUL1 = U 'UL1 (1) = 0.06 × 0.5 + 0.04 × 0.3 + 0.9 × 0 = 0.042
EUL 2 = U 'UL 2 (1) = 0.03 × 0.5 + 0.21× 0.1 + 0.02 × 0.3 + 0.74 × 0 = 0.042
The absolute value of the EUL is 0.042⋅100 = 4.2 MW for both
generating systems. Multiplying this index by T, the system operating time
considered, one can obtain the EUE.
In complex power generation systems consisting of different generating
units each unit besides its total failure state (with zero capacity) and full
operation state (with nominal capacity) can have several intermediate states
due to failures and/or maintenance and inspection actions leading to partial

unit incapacitation (Billinton and Li 1992). The unit capacity in the
intermediate states is less then its nominal capacity, but greater than zero.
Therefore each generating unit should be considered as a multi-state
element with number of states greater than two.
The probability of each state can be obtained based on failure statistics
and maintenance schedule. Having the possible generating capacity levels
and the corresponding probabilities for each unit one obtains the pmf of
its random capacity. For the generating system consisting of different
generating units with determined capacity distributions one can easily
obtain the total capacity distribution (pmf) using the universal generating
function method.
Consider a generation system consisting of n different generating units.
Since the total system capacity is equal to the sum of capacities of the
generating units:
n
G= ∑ Gn , (3.14)
j =1
One can obtain the u-function representing the pmf of the entire system
capacity UG(z) applying the following operators over u-functions UGj(z)
representing the pmf of the units’ capacities:
U G ( z ) = U G1 ( z ) ⊗U G 2 ( z ) ⊗ ... ⊗U G n ( z ). (3.15)
+ + +
Example 3.2
Consider a generation system consisting of three generating units with
variable random capacity (the capacity variations are caused by failures and
maintenance/inspection actions). The system should supply a variable two-
level load L. The high level (day) load is 260 MW and has the probability
0.6; the low level (night) load is 180 MW and has the probability 0.4. The
distributions of unit capacities and of the load are presented in Table 3.2.
The u-functions representing the capacity distribution of the generating
units (the pmf of random variables G1, G2and G3) take the form
UG1(z) = 0.02z0+0.98z50, UG2(z) = 0.02z0+0.02z50+0.96z100,
UG3(z) = 0.03z0+0.01z50+0.02z100+0.94z150.
Table 3.2. Distributions of the unit generation capacities and the load
State 1 State 2 State 3 State 4
First generating g (MW) 50 0 - -
unit p 0.98 0.02 - -
Second g (MW) 100 50 0 -
generating unit p 0.96 0.02 0.02 -
Third generating g (MW) 150 100 50 0
unit p 0.94 0.02 0.01 0.03
Load l (MW) 240 110 - -
q 0.6 0.4 - -
The u-function representing the capacity distribution of the entire

generation system (the pmf of random variable G) can be obtained as
UG(z)=UG1(z) ⊗ UG2(z) ⊗ UG3(z)=(0.02z0+0.98z50)

+ +
0 50 100
⊗ (0.02z +0.02z +0.96z ) ⊗ ( 0.03z0+0.01z50+0.02z100+0.94z150)
+ +
=(0.02z +0.98z )(0.02z +0.02z50+0.96z100)(0.03z0+0.01z50+0.02z100

0 50 0
+0.94z150)=10-4(0.12z0+6.04z50+13.72z100+293.88z150+289.84z200
+552.88z250+8843.52z300).
The expected steady-state capacity of the generation system can be
obtained directly from the u-function representing the pmf of the system
capacity:
E(G)=U'G(1)=10-4(0.12×0+6.04×50+13.72×100+293.88×150
+289.84×200+552.88×250+8843.52×300)=289.5
The u-function corresponding to the pmf of the LLF are obtained using
the composition operator:
U LLF 1 ( z ) = U G ( z ) ⊗ U L ( z )
LLF
= 10-4 (0.12 z 0 + 6.04 z 50 + 13.72 z100 + 293.88 z150 + 289.84 z 200

+ 552.88 z 250 + 8843.52 z 300 ) ⊗ (0.6 z 240 + 0.4 z110 )
LLF
= 0.6 ⋅ 10-4 (0.12 z1(0< 240) + 6.04 z1(50< 240) + 13.72 z1(100< 240) + 293.88 z1(150< 240)
+ 289.84 z1( 200< 240) + 552.88 z1( 250< 240) + 8843.52 z1(300< 240) )
+ 0.4 ⋅ 10-4 (0.12 z1(0<110) + 6.04 z1(50<110) + 13.72 z1(100<110) + 293.88 z1(150<110)
+ 289.84 z1( 200<110) + 552.88 z1( 250<110) + 8843.52 z1(300<110) )
= 0.6 ⋅10 -4 (0.12 z1 + 6.04 z1 + 13.72 z1 + 293.88 z1 + 289.84 z1

+ 552.88 z 0 + 8843.52 z 0) ) + 0.4 ⋅10 - 4 (0.12 z1 + 6.04 z1 + 13.72 z1
+ 293.88 z 0 + 289.84 z 0 + 552.88 z 0 + 8843.52 z 0 )
= 0.6 ⋅10 - 4 (603.6 z1 + 9396.4 z 0 ) + 0.4 ⋅10 - 4 (19.88 z1 + 9980.12 z 0 )
= 10 - 4 (370.112 z1 + 9629.888 z 0 ).
The system LOLP (expected LLF) is
LOLP = U 'LLF (1) = 0.0370
The u-function corresponding to the pmf of the unsupplied load function

is obtained using the composition operator:
U LLF ( z ) = U G ( z ) ⊗ U L ( z )
UL
= 10 (0.12 z + 6.04 z 50 + 13.72 z100 + 293.88 z150 + 289.84 z 200

-4 0
+ 552.88 z 250 + 8843.52 z 300 ) ⊗ (0.6 z 240 + 0.4 z110 )

UL
-4 240 190
= 0.6 ⋅ 10 (0.12 z + 6.04 z + 13.72 z140 + 293.88 z 90
+ 289.84 z 40 + 552.88 z 0 + 8843.52 z 0 ) + 0.4 ⋅ 10-4 (0.12 z110 + 6.04 z 60
+ 13.72 z10 + 293.88 z 0 + 289.84 z 0 + 552.88 z 0 + 8843.52 z 0 ).
The Expected Unsupplied Load is

EUL = U 'UL (1) = 0.6 ⋅ 10-4 (0.12 × 240 + 6.04 × 190 + 13.72 × 140
+ 293.88 × 90 + 289.84 × 40) + 0.4 ⋅ 10-4 (0.12 × 110 + 6.04 × 60
+ 13.72 × 10) = 2.489 MW.
3.2 Evaluating generation system reliability with respect

to unit commitment and operating constraints
There is a growing need within electric utility system operation to refine

approaches to assessing reliability parameters as loss of load probability
(LOLP) and expected unsupplied energy (EUE). These parameters are
greatly affected by the system operation policy requirements such as
spinning reserve requirements, the minimum unit in operation times, the
minimum unit out of operation times and unit starting times. Account of
those requirements renders the methods like those presented in (Baleriaux
et al. 1967) and (Booth 1972) convolution impossible. In our opinion
Monte-Carlo simulation of the generating units' availability is preferable
for calculating the reliability parameters when considering the system

operation policy. An important fact is that uncertain properties of a
generating unit are dependent on its current state. Therefore the use of
Markov transition matrices in approximation of electric utility system
operation is more adequate than using the partial forced outage
probabilities. Obviously, both unit commitment plans and forced outages of
generating units have a direct effect on the value of available system
capacity.
Therefore the main objectives of the presented method are:
• to suggest a Monte-Carlo based reliability evaluation model taking into
consideration the system operation policy and constraints and using the
transition probabilities;
• to measure the sensitivity of reliability indices with respect to variations
in generating units parameters;
• to demonstrate the utilization of the model in assessing the effect of the
system operation policy on the system reliability when using real utility
provided data;
• to develop the model that can be used for long range planning purposes.
The suggested RMUCM (Reliability Modeling with Unit Commitment
by Monte-Carlo) algorithm based on the Monte-Carlo simulation provides
a computationally effective tool for system reliability indices calculation
and sensivity analysis with respect to system operation policy and
generating units' parameters. The reminder of this section details the
approach first sketched in (Gurevich et al. 1995).
The existing tools for evaluation of the generation system operation
reliability are basically different with respect to the method of representing
the system load levels. For example, models such as EGEAS (Fleck and
Charny 1984), Prosceen (Ashworth 1992), WASP (Jenkins and Joy 1974),
Maintenance Scheduling Optimization Program (Gurevich et al. 1992) rely
on the load duration curve. In turn, models such as POWRSYM (Babb
1983), BENCHMARK and PROMOD make use of chronological hourly
load. Both approaches have advantages and limitations. Load duration
curve representation is usually used in long range planning models where a
rough evaluation of the system reliability is performed for various
combinations of planning alternatives. In this case enormously long
computation time prevents the possibility of increasing the accuracy of
calculations. For the same reason the load duration curves are used in many
maintenance scheduling algorithms.
In turn, the short term planning models, performing chronological
detailed probabilistic simulation of power system operation, are usually
based on traditional Forced Outage Rates (FOR) of the generating units,
thus suffering from the lack of information about transitional probabilities

between different availability states of a generating unit.
Obviously, the Monte-Carlo based approaches enable much better
simulation of the real operation of a power system, thus leading to a much
more precise evaluation of reliability indices (Billinton and Li 1994).
Nevertheless, the limitation of these methods is the considerable compu-
tational effort (proportional to the number of samplings) depending on the
desired accuracy of estimation.
The approach presented here separates the Monte-Carlo sampling
process from the system reliability evaluation. First, the Monte-Carlo
sampling process is performed for some predefined number of time periods
and generating units. Thereafter, the desired reliability calculations (inclu-
ding the system operation simulation) require only a short computational
time, as the major computational effort had already been made at the first
stage. Such an approach allows performing the sensitivity analysis of the
results regarding possible variations in system operation policy and
generating units' parameters.
The reliability evaluation problem consists of the assessment of gene-
ration capacity adequacy, taking into account the equipment failures, the
reduced capacity states and operation requirements and constraints.
The period of time for generation system reliability evaluation (the study
period) can be subdivided into T small time intervals during which the
availability state of the generation system is unchanged. The generation
system is composed of N units. Each unit can be found in one out of a set
of possible states. Four types of unit states are considered:
1. in operation with operating capacity equal to nominal;
2. in operation with maintenance reduced capacity (or out of operation);
3. in operation with forced outage reduced capacity (or out of operation);
4. out of operation as a result of a maintenance schedule performed prior to
the study period.
Let the state of the generation system at the instant t (t = 1, ..., T) be
represented by the vector x(t) = (x1(t),x2(t,...,xN(t)) , where xi(t) is the state
of unit i. The trajectory of all the states of generation system during the
study period is denoted by X and is further called the scenario of generation
system reliability evaluation. The set of all possible scenarios x comprises
the scenario space X.
For every scenario x∈ X there is an associated probability P(x). Finally,
let LOLP(x) and EUE(x) represent loss of load probability and expected
unsupplied energy for a given scenario x. We have
T
LOLP(x) = ∑ LOLP ( x(t )) , (3.16)
t =1
and
T
EUE(x) = ∑ EUE ( x(t )) . (3.17)
t =1
Here EUE(x(t)) represents the energy not supplied at the instant t for
scenario x and
⎧1 if EUE ( x(t )) > 0
LOLP(x(t)) = ⎨ (3.18)
⎩ 0 otherwise
The expected value of loss of load probability is calculated as:
LOLP = ∑ LOLP( x) ∗ P( x) , (3.19)
x∈ X
and the expected value of unsupplied energy is calculated as :

EUE = ∑ EUE ( x) ∗ P( x) , (3.20)
x∈ X
In the Monte-Carlo method it is presumed that each scenario has the

same probability. If LOLP is the expected value of LOLP, and EUE is the
expected value of EUE then these values are estimated as
1 M
LOLP = ∑ LOLP( x m ) (3.21)
M m =1
and
1 M
EUE = ∑ EUE ( x m ) , (3.22)
M m =1
where LOLP is the estimate of loss of load probability, EUE is the

estimate of expected unsupplied energy, M is number of samplings, xm is
mth sampled value.
It is important to observe that LOLP and EUE are not the real expected
values of LOLP and EUE but their estimates. As x, LOLP(x) and EUE(x)
are random variables; the estimators LOLP and EUE which are the
averages of M samples of random variables LOLP(x) and EUE(x), are also
random variables. The uncertainties about the estimates are given by the
variance of the estimators:
V( LOLP ) = V(LOLP) / M (3.23)
and
V( EUE ) = V(EUE) / M, (3.24)

where variances V(LOLP) and V( EUE) are estimated as
1 M
V(LOLP) = ∑ ( LOLP ( x m ) − LOLP)2 (3.25)
M m =1
and
1 M
V(EUE) = ∑ ( EUE ( x m ) − EUE ) 2 (3.26)
M m =1
These expressions indicate that the uncertainty of the estimates depends

on the variances of LOLP and EUE and is inversely proportional to the
number of samplings. This uncertainty is often represented as a relative
uncertainty or coefficient of variation:
β LOLP = LOLP / LOLP (3.27)

and
β EUE = EUE / EUE (3.28)

The Monte-Carlo based reliability evaluation scheme can then be
summarized in the following steps:
1. initialize the number of samples M = 0;
2. obtain a scenario xm∈ X by probability simulation, update M;
3. calculate the reliability parameters LOLP and EUE for the sampled
scenario xm;
4. estimate the expected values of the reliability indices LOLP and EUE ;
5. calculate the uncertainty of the estimates;
6. if the uncertainty is acceptable, stop the simulation process. Otherwise
return to step 2.
The expressions defining the coefficients of variation can be used to
estimate the required number of samplings for a given accuracy β.
Substituting, for V( LOLP ) and V( EUE ), their expressions we obtain
β LOLP = V ( LOLP ) / M / LOLP (3.29)
and
β EUE = V ( EUE ) / M / EUE (3.30)
Rewriting these expressions with respect to M gives

⎧⎪ V ( LOLP ) V ( EUE ) ⎫⎪
M = max ⎨ , ⎬ (3.31)
⎪⎩ ( β ∗ LOLP ) ( β ∗ EUE ) 2 ⎪⎭
2
This expression illustrates some advantages and limitations of the

Monte-Carlo method:
• Computational effort is not much affected by the system size: for this
reason Monte-Carlo methods are very suitable for handling the complex
process, such as generation system operation simulation;
• Computational effort is affected by the variance of the estimated
reliability parameters;
• Computational effort is substantially affected by the required accuracy.
3.2.1 Monte-Carlo sampling of the system availability states
Scenario of the generation system capacity availability consists of N

scenarios of all generating facilities. Each scenario of a generating unit is
obtained by probability simulation. There is a planned maintenance outage
schedule for each study period. The maintenance schedule is considered as
an input for probability simulation of generation system operation and it is
the same for all the samples. At the end of the maintenance period the
generating unit available capacity is set equal to nominal.
The set S of unit’s availability states contains three subsets:
• S1 consists of only one state when the unit's available capacity is equal to
nominal;
• S2 consists of the states with the unit’s available capacity reduced (or at
full outage) as a result of a failure;
• S3 consists of the states when the unit’s available capacity is reduced (or
totally zeroed) as a result of a maintenance which was planned prior to
the unit commitment fulfillment.
There are many small maintenance actions (with durations less than 48
hours) which are not planned in advance. The necessity of those actions has
emerged during the generation system operation. It is possible to postpone
such actions and perform it during the convenient hours (night hours as a
rule). Therefore the twenty-four hours are subdivided into the two time
subperiods: night PT1 and day PT2. The statistical analysis shows that those
maintenance periods are distributed uniformly during each subperiod.
The sets S2 and S3 consist of K' states. Each state is defined by a capacity
level from 0 to nominal (not including nominal as it belongs to S1). Thus, a
generating unit can be put in one of the K=2*K'+1 states. For the
probability simulation of generation system operation it is assumed that
transition from one availability state to another may only depend on the
assumed current state and belong to the corresponding time subperiod PTτ.
In such a way a generating unit state transitions can be modeled by Markov
process.
Let us denote the probability of transition of unit n to state j from state i
in time subperiod PTτ by pnτij. For each unit n there are two transition
matrices Pnτ corresponding to the subperiod PTτ:
⎛ pnτ 11 pnτ 12 … pnτ 1K ⎞
⎜ ⎟
⎜p pnτ 22 … pnτ 2 K ⎟
Pnτ = ⎜ nτ 21 (3.32)
… … … … ⎟
⎜ ⎟
⎜p pnτK 2 … pnτKK ⎟⎠
⎝ nτK1
These matrices are known as the stochastic transitional probability
matrices. It should be noted that the sum of all the probabilities in each row
of the matrix must equal 1 since each row i represents all the possible
transitions to the other states in a particular instant of time given that it is in
the state i.
For a class of Markov chain possessing the ergodic property, the state
probabilities will converge to a vector of steady-state probabilities
pnτ = (pnτ1 pnτ2 … pnτK) (3.33)
such that
pnτ * Pnτ = pnτ. (3.34)
This matrix equation together with
K
∑ pnτk = 1 (3.35)
k =1
determine the vector pnτ.

To sample a generating unit state in the initial instant of time a value of x
is sampled using the [0,1] uniform distribution. Thereafter, the index k(x) is
found to satisfy the inequality
k ( x ) −1 k ( x)
∑ pn1k < x ≤ ∑ pn1k (3.36)
k =1 k =1
(it is assumed that the initial instant of time coincides with the beginning of
twenty-four hours daily period, and consequently belongs to the first time
subperiod PT1). The initial state k0 of the generating unit is equal to k(x) To
sample a generating unit state in a certain instant of time the same sampling
procedure as in the initial instant is performed first to determine the value
x. Next the index k(x) is found to satisfy the inequality
k ( x ) −1 k ( x)
∑ pnτk s k < x ≤ ∑ pnτk s k (3.37)
k =1 k =1
in the corresponding subperiod PTτ for the current state ks. The next unit
state kn is equal to k(x). Execution of the sampling procedure for each
instant of time during the study period results in the unit availability
scenario.
The set Tn of instants at which kn≠ks and the available unit capacity has
changed divides the study period into time intervals. The state of the unit,
regarding the unit commitment during each interval, does not change. Set
Tn contains the set Tn of instants at which kn∈ S2 (the instants of failures)
and the available unit capacity decreased. The set Tn divides the study
period into time intervals so that the unit state can vary during these
intervals but these variations are known beforehand and, therefore, it is
possible to take them into account in unit commitment procedure.
Joining up all generating unit availability scenarios leads to the entire
generation system availability scenario. The set
N
T= ∪ Tn (3.38)
n =1
divides the study period into time intervals so that the variations of the
system state during each time interval are known from the beginning (there
are no failures during this interval). Consequently, it is possible to fulfill
the unit commitment for such an interval entirely.
Now we have a sequence of the generation system operation scenarios.
Each scenario consists of integration of generating unit operation scenarios
and subdivision of the study period into the time intervals without failures.
3.2.2 Implementing unit commitment
The purpose of unit commitment is to define, for each unit, the on line
intervals during the unit commitment period to meet the load demand and
to satisfy the generation system operation policy. The unit commitment
period is subdivided into time intervals of the same length. This length is
equal to the minimum time in order to start up an emergency unit (as a rule
a jet turbine). Load is represented by peak loads at these time intervals.
The spinning reserve requirements are given in percent of load demand
or in MW and may vary during the study period. It should be noted that
sometimes it is impossible to satisfy the spinning-reserve requirements
because of the generating units capacity shortages. These constraints can be
written in the form
∑ OC jt ≥ LOADt + SRt (3.39)
j∈U t
where j is index of unit, t is time instant, OCjt is available capacity of unit j

at instant t, LOADt is load demand at instant t, SRt is spinning reserve
requirement at instant t, Ut is set of committed units at instant t.
The generation system operating constraints include:
• unit minimum uptime requirement according to which a generating unit
is started up only if it is needed for more than a specified minimum
number of hours of continuous operation;
• unit minimum down time rule. which does not permit a generating unit
to be started up if it was out of operation less than a specified minimum
number of hours;
• unit minimum start time, which specifies the minimum number of hours
needed for bringing the unit from down to operating condition;
• unit must-run rule, which requires a specified generating unit to be
operated all the time when it is available.
To perform the unit commitment the units are put into the priority queue
and brought into operation according to their position in the queue. Must-
run units (units that can never be shut down) are put first in the priority
queue. Other units are ordered according to increasing production cost,
decreasing minimum up/down time and increasing start up cost. Peaking
units with high production cost and short minimum up/down time are
located last in the priority queue.
To describe the unit commitment procedure the following notation is
introduced:
j
TMDN unit’s minimum down time,
j
TMUP unit’s minimum up time,
TSTj unit’s minimum start time,
SLjt unfulfilled system demand faced at the instant t,
Τ duration of the unit commitment horizon,
τ0 time during which unit j has been on-line (if τ0>0) or off-line
(if τ0 <0),
τs time from switching-on the unit j if it has been initially off-line

(τ0 <0) or 0 if the unit continues to be switched-off.
For each generating unit which is available and does not have the must-
run status the first free-decision instant t0 is defined as follows:
⎧ max{T j − τ + 1,1} if τ 0 > 0
MUP 0
⎪⎪ j
t0 = ⎨ max{TST − τ s + 1,1} if τ 0 < 0 and τ s > 0 (3.40)
⎪max{T j j
⎪⎩ MDN + τ 0 + 1, TST + 1,1} if τ 0 < 0 and τ s = 0
If τ0>0 then unit j is on line in interval [1,t0) and off line otherwise.
Based on the commitment priority rule, unit j should be committed at a
free-decision instant t∈ [t0,T] if there is unsupplied system demand at this
instant ( SLjt > 0 ). As a result, each free-decision instant t must be either of
the following:
• An on line instant (SLjt > 0 ),
• An unclassified instant ( SLjt ≤ 0 ) , for which commitment status has not
yet been determined
Let Ik, k = 1,...,K represent the closed interval k of consecutive
unclassified instants contained in the free decision interval [t0,T], and let Tk
represent the duration of interval Ik. These unclassified intervals are
numbered according to their chronological orders. Two adjacent
unclassified intervals are separated by an on line interval. Let Ik = [tk1,tk2] ,
then
Tk = tk2 - tk1 + 1. (3.41)
Let τk represent the duration of interval when unit j was on-line (τk >0) or
off-line (τk<0) just before the instant tk1. Based on τk, the tentative
commitment strategy associated with unclassified interval Ik can be
estimated as follows:
j
• If 0 < τk < TMUP - Tk then unit will be set on-line during the whole
interval Ik;
j j
• If 0 < τk < TMUP and τk > TMUP - Tk then determine
j j
tk1 = tk1 + (TMUP − τ k ) , Tk = Tk − (TMUP −τ k )
j
− If Tk < TMDN then the unit will be set on line during the whole
interval Ik;
j
− If Tk ≥ TMDN then the unit will be set on line during the interval
[tk1, tk1 ) and off-line during the interval [ tk1 ,tk2]
j j
• If τk ≥ TMUP and Tk < TMDN then the unit will be set on-line during the
whole interval Ik;
j j
• If τk ≥ TMUP and Tk ≥ TMDN then the unit will be set off-line during the
whole interval Ik;
If interval Ik is the first one (k = 1) and it begins at t0 then the following

rules are added to the commitment strategy:
j
• If τs > 0 and T1 ≤ TMUP then the unit will be set on-line during the whole
interval I1 ;
j j j
• If τs > 0 and T1 > TMUP then determine t11 = t11 + TMUP , T1 = T1 − TMUP
j
− If T1 < TMDN then the unit will be set on line during the whole
interval I1 ;
j
− If T1 ≥ TMDN then the unit will be set on line during the interval
[t11, t11 ) and off-line during the interval [ t11 , t12].
− If τs = 0 and τ1 < 0 then the unit will be set off line during the
whole interval I1.
It should be noted that the commitment strategy associated with the
unclassified interval Tk can be determined if τk is known. Since τk (and τs , if
k=1) can be determined from js unit known commitment status prior to the
interval Ik, the commitment strategy associated with the interval Ik can be
evaluated. Based on τk and the commitment strategy associated with the
interval I1, τk+1 can be easily determined. Thus, the commitment strategy
associated with each unclassified interval can be recursively estimated.
If there are not enough units to meet the system demand at each instant,
the sum of all the positive parts of differences between the total available
generation system capacity and the system demand at each instant during
the unit commitment horizon is added to the expected unsupplied energy
and the total number of these instants contributes to the loss of load
probability. When a failure happens at the last instant of the unit
commitment horizon, then capacity drop is subtracted from the committed
capacity if it was committed. If the remaining committed capacity does not
meet the system demand then their difference is also added to the expected
unsupplied energy and the number of loss of load instants is increased by
unity .
The model described above provides a means for rapid evaluation of the
influence of the system operation rules on reliability indices. It may
become an efficient tool for choosing the proper system operation policy
supporting the reliability requirements without additional capacities.
Example 3.3
Using a very simple example we show how to model power system
operation for obtaining a statistically reliable assessment of the loss-of-load
probability and the unserved energy. The random nature of the forced
outages is described by the Markov chains combined with the Monte-Carlo
simulations of the unit availability.
The generating system is formed of only five units detailed in Table 3.3
Table 3.3. Parameters of generating units
Nominal Minimum Min Min Start

Unit Capacity Capacity UpTime DownTime Time
(Mw) (Mw) (minutes) (minutes) (minutes)
1 500 250 480 360 120
2 200 100 360 240 60
3 300 150 240 60 30
4 100 50 60 30 15
5 20 10 10 0 0
The units are sorted in the ascending order of their power generating
cost, that is, the first unit produces the cheapest energy. This sorting
defines the merit order for unit commitment.
We also suppose that each unit can be in either one of three different
states: full unavailability (i=1), 50% availability (minimum capacity) (i=2)
and full availability (i=3). The (3x3) probability transition matrices are
presented in Table 3.4, where Pni stands for the probability of the unit
transition from the n-th to i-th state every 15 minutes. The diagonal entries
of the matrices dominate the off-diagonal ones due to the unit ability of
preserving its current state. The transition diagram, corresponding to the
first matrix is shown on Fig. 3.2.
Table 3.4. The transution probabilities matrices

Unit 1 Unit 2 Unit 3
Pni i=1 i=2 i=3 i=1 i=2 i=3 i=1 i=2 i=3
n=1 95% 3% 2% 92% 5% 3% 85% 7% 8%
n=2 1% 97% 2% 4% 91% 5% 7% 87% 6%
n=3 2% 4% 94% 6% 4% 90% 6% 10% 84%
Table 3.4. (cont.)
Unit 4 Unit 5
Pni i=1 i=2 i=3 i=1 i=2 i=3
n=1 98% 1% 1% 93% 3% 4%
n=2 2% 97% 1% 4% 91% 5%
n=3 3% 2% 95% 5% 3% 92%
0.02
0.95 0.94
0.97
0.03 0.02
500 Mw 250 Mw 0 Mw
0.01
0.04
0.02
Fig. 3.2. Transition diagram of generating unit

For further simplicity, we assume that the load demand (LD) and
spinning reserve (SR) are constant during 3 hours: LD = 900 Mw, SR = 50
Mw. The system operation should be organized in such a way as to provide
(LD+SR) = 950 Mw of available capacity all the time at minimum cost.
The installed capacity of the system (500+200+300+100+20) = 1120 Mw
exceeds this requirement, so the unit commitment/optimal dispatch
scheduling would be a very simple if there were not unforeseen forced
outages. At specific moments they can result in the system failure to
provide a required demand. In such a case one should discriminate between
impossibility to satisfy the load demand or only the spinning reserve
requirement. The first case results in a specific amount of the unserved
energy while the latter is only potentially dangerous. Depending on the
future system state, it may also lead to the unserved energy at the
subsequent moments or show no impact on the consumers. Both cases are
detailed in the rest of the paragraph.
Table 3.5 presents randomly generated availability of each unit for the
three hours simulation period with the 15-minutes resolution.
Table 3.5. Availability of generating units

Time Units
(Minutes) 1 2 3 4 5 Total
0-15 500 200 300 100 20 1120
15-30 500 100 300 100 20 1020
30-45 500 100 300 100 20 1020
45-60 500 100 300 100 20 1020
60-75 500 100 300 100 20 1020
75-90 500 200 300 100 20 1120
90-105 500 200 300 100 20 1120
105-120 500 200 300 100 20 1120
120-135 500 200 300 100 20 1120
135-150 500 200 150 100 20 970
150-165 500 100 150 100 20 870
165-180 500 100 150 100 20 870
180-195 500 200 300 100 20 1120
195-210 500 200 300 100 20 1120
210-225 500 200 300 100 20 1120
225-240 500 200 300 100 20 1120
For reader convenience the partial and full unavailability of the units are
boldfaced.
At first glance, the total availability of the units permits to provide
LD+SR at each moment. However, in actual practice, the more expensive
units (4 and 5 in our example) are shut down, whenever possible, to
minimize the energy generation cost. This policy may lead to the SR
shortage or, more severely, to the unsupplied energy, when either of the
basic units 1-3 is not fully available due to an unexpected failure. Indeed,
the unit 4 cannot be started immediately (because of its relatively long start
up time) while the quickly started unit 5 has too small capacity for
compensating even a partial unavailability of the basic units.
Tables 3.6 and 3.7 display the corresponding dispatch and spinning
reserve contributions of the units.
At the first moment all the units are fully available. The generation of the
required LD = 900 Mw is provided by nominal loading of the two cheapest
units and partial loading (200 Mw) of the third unit. The rest of its available
capacity (300-200) = 100 Mw covers the required SR = 50Mw. The
expensive units 4 and 5 are not started up at all. At the moment (Time =
15) unit 2 drops down to only 50% of its nominal capacity 100 Mw), so
that the total available on-line capacity at this moment is (500+100+300) =
900 Mw. This is sufficient only to satisfy the load demand (900 Mw) with
no spinning reserve contribution.
Table 3.6. Dispatch of generating units during 3 hours.

Time Units
0-15 500 200 200 900
15-30 500 100 290 10 900
30-45 500 100 250 50 900
45-60 500 100 250 50 900
60-75 500 100 250 50 900
75-90 500 200 150 50 900
90-105 500 200 200 900
105-120 500 200 200 900
120-135 500 200 200 900
135-150 500 200 150 20 870
150-165 500 100 150 100 20 870
165-180 500 100 150 100 20 870
180-195 500 200 150 50 900
195-210 500 200 150 50 900
210-225 500 200 200 900
225-240 500 200 200 900
Table 3.7. Spinning reserve contribution of generating units
Time Units
0-15 0 0 100 50
15-30 0 0 10 10 20
30-45 0 0 50 50 100
45-60 0 0 50 50 100
60-75 0 0 50 50 100
75-90 0 0 150 50 200
90-105 0 0 100 100
105-120 0 0 100 100
120-135 0 0 100 100
135-150 0 0 0 0 0
150-165 0 0 0 0 0
165-180 0 0 0 0 0
180-195 0 0 150 50 200
195-210 0 0 150 50 200
210-225 0 0 100 50
225-240 0 0 100 50
Unit 4, though available, cannot be started up immediately due to its

relatively long start-up time. For this reason, the only possible reaction is to
start up unit 5 with practically zero start-up time at minimum loading level
(10Mw) and to reduce loading of unit 3 to 290 Mw. This operation
contributes only 20 Mw to the spinning reserve.
The shortage (50-20) = 30Mw of spinning reserve forces the dispatcher
to start up unit 4 which becomes available after 15 minutes, which are
needed for the start-up, at the moment (Time = 30). Unit 4 is loaded at the
minimal level of 50 Mw and unit 3 is loaded at the level of 250 Mw.
Spinning reserve contribution will be 100 Mw.
It continues to work up to the moment (Time = 75), when unit 2 restores
its full availability. For saving purposes, unit 4 should be shut down.
Nevertheless, because of the minimum up time requirement it will continue
to be on line for another 15 minutes up to the moment (Time = 90). During
this period unit 2 is loaded at the nominal capacity and unit 3 at the level of
150 Mw. Spinning reserve contribution will be 200 Mw.
At the moment (Time = 90) unit 4 is shut down and unit 3 is loaded at
the level 200 Mw. Spinning reserve contribution will be 100 Mw.
At the moment (Time = 135), unit 3 possesses only 50% of its
availability. Again, as before, unit 4 can not be started up immediately.
Instead, the dispatcher performs the quick start of unit 5. This gives totally
(500+200+150+20) = 870Mw, with (900-870) = 30Mw unsupplied.
At the moment (Time = 150) unit 4 is started but unit 2 drops to the level
of 100 Mw. With no regard to ramp rate requirement let us suppose that the
load of unit 4 could be immediately increased to the nominal level.
Therefore, the total available capacity is (500+100+150+100+20) =
870Mw, with (900-870) = 30Mw unsupplied up to the moment Time =
180, when units 2 and 3 will be available.
Due to the minimum down requirements unit 4 is not shut down until the
moment Time = 210. Thus, the load of unit 2 is increased to the nominal
capacity, the load of unit 4 is decreased to the 50Mw, unit 5 is shut down
and the load of unit 3 is not changed. Spinning reserve contribution is then
200 Mw.
At the moment Time = 210 unit 4 is shut down and unit 3 is loaded at the
level 200 Mw. Spinning reserve contribution is equal to 100 Mw.
In this example we have 45 minutes of loss of load = 0.75LOLH
(Time135-180) and the unsupplied energy (30Mw*0.75) = 22.5MWh.
3.3 Optimal reserve management for power generation

systems
The purpose of power generating systems is to provide reliable and

economical electrical energy to their customers. Back up resource or
energy, which is designated as reserve, is commonly used in generating
systems in order to provide adequate supply under unexpected system
operating conditions such as generating system failures and load growth. A
generating system can plan its own reserve and can also share reserve with
other generating systems according to their reserve contracts. The reserve
structure of a generating system should be determined based on the balance
between the required reliability and the cost of reserve. The objective of
reserve management for a generating system is to schedule the reserve at
the minimum total system reserve cost while maintaining a required level
of supply reliability to its customers (Ding, Wang, Lisnianski 2006).
A generating system usually consists of generating units (GUs) with
different capacity, cost and availability. Different generating systems are
usually connected together through supply network to form a complex
generating system (CGS) as shown in Fig. 3.3.
The supply network is designated as transmitting system (TS). The
objective of the interconnection is to share the reserve. A GS usually signs
the reserve agreements with other GSs to increase its reliability and to
reduce the cost of reserve. A CGS can operate in different states or
performance levels due to random failures. The performance levels of a GS
in the CGS are usually determined by the number of units, capacity and
availability of each unit, transmitting system and reserve agreements with
other GSs. The reserve management of a generating system in a CGS is,
therefore, a complex optimization problem. A CGS and its GSs and TSs are
multi-state systems (Lisnianski and Levitin 2003) that can be represented
using the corresponding equivalents (Wang and Billinton 2003, Billinton
and Wang 1999).
Generating System 1 Generating System 2
GU1,1 GU2,1
... ... ... ...

Transmitting System
GU1,N GU2,N
... ...
GUM-1,1 GUM,1
... ... ... ...
GUM-1,N GUM,N
Generating System M-1 Generating System M
Fig. 3.3. Complex generation system
3.3.1 Multi-state system models
In order to determine the optimal reserve structure for a specific GS in a

CGS, this GS, other GSs (serving as reserve providers) and the network
between the GS and providers are represented as multi-state elements
(multi-state generating system MSGS, multi-state reserve providers MSRPs
and multi-state transmission system respectively). The model of a CGS for
the reserve optimization is shown in Fig. 3.4.
Assume that MSGS has K g states and each state i g is associated with
available generating capacity AGig . These states can be divided into two
sub-sets: the normal states (NS) and the contingency states (CS). For the
normal states, AGi is greater or equal than a load of the MSGS. For the
g
contingency states, AGi is less than a load of the MSGS. We assume that
{ 1,2,..., m }∈NS, and { m + 1,..., K g }∈ CS.
g
g g
MSRP1 MSTS1
Reserve
MSRP2 MSTS2 MSGS
…
Reserve
MSRPM MSTSM
Fig. 3.4. Equivalent Model of a CGS for the MSGS
The performance level of MSRP m is characterized by its available

generating capacity ( AGmi that can take K mr values in accordance with its
r
states) and available reserve capacity ARmi . ARmi is determined by AGmi ,

r r r
by the contractual reserve capacity CRm , and by the types of contracts.

There are two types of contracts: firm and infirm.
For the firm contract, ARmir = CRm if AG mi ≥ CRm and ARmir = AGmi if
r r
AG mi ≤ CRm so that
r
ARmir = min{ AGmir , CRm } (3.42)
For the infirm contract, CRm is not guaranteed. It depends on the load
level of MSRP m for interval n Lnm . ARmi is defined as follows:
r
ARmir = min{max( AG mir − Lnm , 0), CR m } (3.43)
The performance level of MSTS h is represented by its available

transmission capacity ATmi that can take K mt values in accordance with its
t
states.
The number of different states for the entire CGS is:
M M
K = K g ∏ K mr ∏ K ht (3.44)
m =1 h =1
If all the elements in the system are independent, the probability of state
i for the entire CGS is determined by the state probabilities of MSGS,
MSRPs and MSTSs associated with state i:
M M
pi = pi g
∏ p mi ∏ p mi
r t
(3.45)
m =1 m =1
When the CGS stays in state i, the performance level wi of the MSGS
after considering the assistance from the MSRPs can be determined based
on AGi , ARmir and ATmit .
g
The capacity distribution of the MSGS is determined by pi and wi . For a

CGS which consists of a large number of elements, determining
performance level of the MSGS is very resource consuming. The
computational burden can be reduced by considering only the states in the
CS of the MSGS.
The variable load is represented by a load duration curve (LDC), which
is the rearrangement of time varying loads in descending order. The LDC is
partitioned into different time periods Tn ( n = 1,..., N ) with constant load
level Ln for each time interval Tn.
Two reliability indices used for a MSGS are the MSGS unavailability or
the lost of load probability (LOLP) and the expected energy not supplied
(EENS):
∑∑ Pr{L }
N K
LOLP = n
> wi ⋅ Tn T (3.46)
n =1 i =1
∑∑ Pr{L }
N K
EENS = n
> wi ⋅ ( Ln − wi ) ⋅ Tn (3.47)
n =1 i =1
Problem formulation
The reserve contract between the MSGS and MSRP m is determined by the
contract type, CRm , per unit costs of reserve capacity and of the utilized
reserve provided by MSRP m ( ε Cm and ε Um respectively), and also by the
reliability of MSRP m and MSTS m. The objective of reserve management
for a MSGS is to determine the optimal reserve structure, which includes
the reserve contracts with MSRPs and the utilization order of MSRPs,
under the minimum total reserve cost while satisfying the system reliability
requirement. The total reserve cost includes the reserve capacity cost CC
and the expected reserve utilization cost EC. The reserve optimization
problem can be summarized as:
M
Min ∑ (CCm + ECm ) (3.48)
m =1
CCm = ε Cm CRmT (3.49)

where
N K
n
ECm = ∑ ∑ piε UmURmi Tn , (3.50)
n =1i =1
subject to the following constraints.

The reserve constraints:
n
URmi ≤ CRm , (3.51)
n
URmi {
≤ min ARmir , ATmit ; } (3.52)
The reliability constraints:

LOLP ≤ LOLP spec , (3.53)
EENS ≤ EENS spec , (3.54)

where LOLP spec and EENS spec are the required levels of the LOLP and
EENS.
The load balance constraint takes the form
M
AGi g + ∑UR
m =1
n
mi = Ln − LLni . (3.55)
Load curtailment LLni is utilized when the total generating and reserve
capacity cannot meet the load of the MSGS.
UGF-based reliability evaluation

For an MSGS with GN generating units connected in parallel, the u-
function representing its total generating capacity distribution can be
represented using the ⊗ operator over u-functions of GN generating units:
+
[ ]
GN
U MSGS ( z ) = ⊗
+
U G1 ( z ),..., U G N ( z ) = ∏U j ( z) =
j = G1
(3.56)
[ ]= ∑ p
Kg
∏ (1 − A j )⋅ z + A j ⋅ z
GN AGig
0 w j ,1
ig ⋅z
j = G1 i g =1
where A j and w j ,1 are the availability and the nominal generating capacity
of generating unit j respectively.
Notice that the first mG terms in the U MSGS (z ) with the associated states
belonging to NS can be removed from the u-function without changing the
results of reliability indices. In order to simplify the algorithm one can
remove the first mG terms from the u-function and define it as
Kg
AGi g
U MSGS ( z ) = ∑ p ig ⋅ z . (3.57)
i g = m g +1
Although all the generators of a MSRP are also connected in parallel, the
u-functions representing the distribution of the random reserve depend on
the type of the reserve contracts. For MSRP m with Rm N generating units
which are connected in parallel, the u-functions of MSRP m for the firm
and infirm contracts can be represented using the operators ⊗ and ⊗ ,
firm infirm
respectively, over u-functions of Rm N generating units.

For the firm contract
[
U MSRP ( z ) = ⊗ U Rm1 ( z ),..., U Rm N ( z ), CRm =
firm
]
⎡ Rm N ⎤
⎡
( ⎤
)
= ⊗ ⎢ ⊗ U Rm1 ( z ),..., U Rm N ( z ) , CRm ⎥ = ⊗ ⎢ ∏ U j ( z ), CRm ⎥
firm ⎣ firm ⎦ firm ⎢⎣ j = Rm1 ⎥⎦
[( ]
⎧⎪ Rm N ⎫⎪ (3.58)
w
)
= ⊗ ⎨ ∏ 1 − A j ⋅ z 0 + A j ⋅ z j ,1 , CRm ⎬
firm ⎪ j = Rm
⎩ 1
⎪⎭
K mr K mr
min( AG mir , CR m ) AR
= ∑ pmir ⋅ z = ∑ pmi ⋅ z mir .
r
i r =1 i r =1
For the infirm contract

[
U MSRP (Z ) = ⊗ U Rm1 ( z ),...,U RmN ( z ), CRm , Lnm =
infirm
]
⎡ RmN ⎤
⎡
( ⎤
)
= ⊗ ⎢ ⊗ U Rm1 ( z ),...,U RmN ( z ) , CRm , Lnm ⎥ = ⊗ ⎢ ∏U j ( z ), CRm , Lnm ⎥
infirm⎣infirm ⎦ infirm⎣⎢ j = Rm1 ⎦⎥
(3.59)
⎧⎪ RmN
[( w
) ⎫⎪
= ⊗ ⎨ ∏ 1 − A j ⋅ z 0 + A j ⋅ z j ,1 , CRm , Lnm ⎬
infirm⎪ j = Rm ⎪⎭
]
⎩ 1
r r
N Km min(max(AGmir − Lnm ,0), CRm ) N Km ARmir
= ∑ ∑ pmir ⋅ z ⋅ Tn / T = ∑ ∑ pmir ⋅ z ⋅Tn / T
n =1ir =1 n =1i r =1
For MSTS m with Tm N transmission lines connecting the MSGS with

MSRP m, the u-function representing the distribution of the transmitting
capacity can be determined using the ⊗ operator:
+
[ ]
Tm N
U MSTS ( Z ) = ⊗ U Tm1 ( z ),..., U Tm N ( z ) =
+
∏U j ( z)
j = Tm1
(3.60)
Tm N
j = Tm1
[
∏ (1 − A j )⋅ z + A j ⋅ z 0 w j ,1
]= ∑ p
K mt
it =1
mit ⋅z
ATmit
where A j and w j ,1 are the availability and the nominal capacity of

transmission line j.
The MSRP m and MSTS m are connected in series. To calculate
the UGF for the series-connected subsystem, the ⊗ operator should be
min
used:
⎡ Kmr AR
Kmt
AT ⎤
⊗ (U MSPR( z),U MSTS( z)) = ⊗ ⎢ ∑ pmir ⋅ z mir , ∑ pmit ⋅ z mit ⎥
min min⎢i =1 ⎥⎦
⎣r it =1
(3.61)
Kmr Kmt
min(ARmir , ATmit )
= ∑ ∑ pmir ⋅ pmit ⋅ z
ir =1it =1
WR
The u-function of the MSGS U MSGS ( z ) with the reserve agreements with
MSRPs and the capacity limits of MSTSs can be now obtained as:
WR
U MSGS { [
( z ) = ⊗ U MSGS ( z ),..., ⊗ U MSRS m ( z ), U MSTS m ( z ) ,..., ]
]}
+ min
(3.62)
[
⊗ U MSRS M ( z ), U MSTS M ( z )
min
WR
From U MSGS ( Z ) we can obtain the LOLP using the following equation
N K
{ }
LOLP = ∑∑Pr Ln > wi ⋅Tn T = ∑ ∑ pi ⋅ 1(Ln − wi )⋅Tn T
N
(3.63)
n=1i=1 n=1i∈CS
The EENS for a given period can be obtained as

N K
{ }
EENS = ∑∑ Pr Ln > wi ⋅ ( Ln − wi ) ⋅ Tn
n=1i=1
(3.64)
∑ Pr{L > wi }⋅ (L − wi ) ⋅ Tn = ∑ ∑ pi max{L − wi , 0}⋅ Tn .
N N
n n n
=∑
n=1i∈CS n=1i∈CS
Implementing the Genetic Algorithm

M
A mixed numerical and binary string of length ∑ Dm is used to encode a
m =1
solution in the GA, where Dm is a number of binary bits encoding the
amount of reserve capacity CRm . A typical chromosome representing three
reserve providers is shown in Fig 3.5.
The first M elements ( a1a 2... a M ) of string represent order of MSRPs
utilization in a contingency state (aj is an order of MSRP j). This substring
should be a permutation of integer numbers from 1 to M.
M
The next ∑ Dm bits represent the contracted reserve capacity CRm for M
m =1
MSRPs. Encoding is performed using different number of bits Dm for each
contracted reserve amount, depending on the desired accuracy. The
decoding procedure obtains the amount of the reserve for each contract as
k
CRm = CRmmin + (CRmmax − CRmmin ) ⋅ (3.65)
2 Dm
−1
where CRm takes the values in the interval [CRmmin , CRmmax ] and k is the
decimal number represented by the binary substring.
Using this encoding algorithm, the solutions for obtaining the reserve
utilization order are within the feasible space. According to the string
represented in Fig. 3.5, MSRP 2 is used first, MSRP 1 is used second, and
so on, up to the point the load is met, or the whole available reserve is used
in a contingency state.
To let the GA look for the solution with minimal total cost without
violating the constraints LOLP ≤ LOLP spec and EENS ≤ EENS spec , the
solution quality (fitness) is defined as
M
F = q ⋅ max(0, LOLP − LOLP spec , EENS − EENS spec ) + ∑ (CCm + ECm ) (3.66)
m =1
where q is a sufficiently large penalty coefficient.

M S R P1 M S R P 2 M S R P 3 M S R P1 M S R P 2 M S R P 3
2 1 3 0 1 ...0 1 1 ...0 0 1 ...1
u tiliz a tio n s e q u e n c e o f c o n tr a c tu a l r e s e r v e
reserve c a p a c ity
Fig. 3.5. GA chromosome structure
Example 3.4
Consider the restructured IEEE-RTS (IEEE Task Force 1979) consisting of
six MSGSs in which the optimal reserve structure should be determined for
the specific MSGS, which has reserve agreements with other five MSGSs.
The MSGS owns 6×50MW hydro units, 5×12MW oil units, 2×155MW
coal units and 2×400MW nuclear units. MSRP1 owns 2×20MW gas
turbine units and 2×76MW coal units. MSRP 2 owns 3×100MW oil units.
MSRP 3 owns 2×20MW gas turbine units and 2×76MW coal units. MSRP
4 owns 3×197MW oil units. MSRP 5 owns 2×155MW coal units and one
350MW coal unit. The MSRPs are connected to the MSGS through tie
lines.
The tie line reliability data (Allan et al. 1986) is used in this paper. We
assume that all the tie lines are identical, each with the capacity rating of
300MW and unavailability 0.00130873. MSTSs 1, 2 and 3 each have one
tie line. MSTSs 4 and 5 each have two tie lines.
The four-step model of the LDC obtained from hourly loads is used. The
peak load of the MSGS is 1253MW. For the MSGS, the load levels Ln and
the total time periods T n of a particular step are shown in Table 3.8. The
load level Ln of a particular step is determined by the average value of
hourly load in the time period of T n .
Table 3.8. The MSGS load level and time period
Ln (MW) 549 754 939 1051

T n (h) 2281 3909 1458 968
All the MSRPs are firm contract providers for the MSGS. The utilization
prices for MSRPs are 25$/MWh, 55$/MWh 75$/MWh, 60$/MWh and
20$/MWh, respectively. The reserve capacity prices for MSRPs are
2.5$/MWh, 5.5$/MWh 7.5$/MWh, 6$/MWh and 2$/MWh, respectively.
The time frame for the contracts is one year. Two examples with different
reliability constraints are investigated in this case.
In example A, the LOLP and EENS constraints for the MSGS are set as
0.01 and 500MWh/yr respectively. In example B, the reliability constraints
are tightened and the LOLP and EENS for the MSGS are 0.0005 and
50MWh/yr respectively.
Table 3.9 shows the total reserve cost and the reserve capacities between
the MSGS and MSRPs for the two examples.
Table 3.9. The total reserve cost and reserve structure for for the first case
Example A Example B
Total Reserve Cost (k$) 10197.0 10896.1
CR MSRP1 (MW) 53.728 70.355
CR MSRP 2 (MW) 1.7895 0.0185
CR MSRP 3 (MW) 0.2981 0.0186
CR MSRP 4 (MW) 6.4949 0.0524
CR MSRP 5 (MW) 374.82 419.02
The capacity reserve cost and expected cost of reserve utilization for
each MSRP are shown in Table 3.10. It can be seen from Tables 3.9 and
3.10 that the reserve cost has increased dramatically when the reliability
requirements were tightened.
Table 3.10. The capacity reserve cost and expected cost of utilized reserve for the
first case
Example A Example B
MSRP CC (k$) CC m (k$)
No. m EC m (K$) EC m ( K$)
1 1172.6 24.7 1536.1 18.2

2 86.0 1.02 0.89 0.0013
3 19.5 0.23 1.22 0.0017
4 340.4 4.05 2.74 0.004
5 6549.4 1999.0 7321.2 2015.7
3.4 Determination of the random generating capacity

distribution for power plant
In this section we consider the evaluation of reliability indices of a

separated power plant. These indices are used for the entire generation
system reliability calculation.
The power plant usually consists of several generating units. Each of the
generating units comprises a complex system of technical components con-
nected in parallel or in series in reliability block diagram sense. In order to
increase the reliability of generating units, redundancy of their auxiliary
facilities (feeders, conveyors, pumps, heaters, condensers, heat exchangers,
air fans, storage tanks etc.) is provided. Failures of some components may
cause a total outage (for example a failure in the boiler) whereas failures of
other components cause only a reduction in the power output of the unit
(for example a failure in one of the feeder pumps). Besides unexpected
failures, planned maintenance of components also affects the unit reliability.
The components are aimed at performing their tasks with a nominal
performance rate. They can usually be represented by a two state model.
When the component h is available its performance Gh is equal to the
nominal performance rate gh and if the component is not available, its
performance equals zero. The pmf of the element performance can be
represented by the following u-function
u h ( z ) = A h z g h + (1 − A h ) z 0 = A h z g h + A h z 0 , (3.67)
where Ah is the availability of component h. The availability of a
component at steady state is calculated by two statistical parameters: the
Mean Time to Failure (MTTF) and the Mean Time to Repair (MTTR).
MTTFh µh
Ah = = (3.68)
MTTFh + MTTRh µ h + λh
where λh and µh are the forced outage rate and the repair rate of component
h respectively, λ = 1/MTTF, µ = l / MTTR.
The unavailability Ah which is also called FOR (Forced Outage Rate),
is calculated as:
λh
Ah = P(Gh = 0) = FOR = . (3.69)
µ h + λh
When component performance is defined as capacity or productivity, the
total capacity G{i, j} of a subsystem containing two independent
components i and j connected in series is equal to the capacity of a

bottleneck component (the component with lower performance):
G{i, j} = min{Gi , G j } . (3.70)
The u-function representing the pmf of G{i, j} can be, therefore, obtained
using the operator ⊗ over the u-functions representing pmf of Gi and Gj.
min
For example the u-function U1,2(z) representing the total performance of a

subsystem consisting of a pair of two-state components with nominal
performances g1 and g2 and availabilities A1 and A2 connected in series
takes the form
U1,2 ( z ) = u1 ( z ) ⊗ u2 ( z )
min
= ( A1z g1 + A1z 0 ) ⊗ ( A2 z g 2 + A2 z 0 )
min (3.71)
min{ g1 , g 2 } 0 0 0
= A1 A2 z + A1 A2 z + A1 A2 z + A1 A2 z
= A1 A2 z min{g1 , g 2 } + (1 − A1 A2 z 0 ).
The u-functions for any series system can be obtained recursively by

consecutively determining the u-functions of pairs of the components. For
example the u-function of a system consisting of four components
connected in a series can be determined in the following way:
u1( z ) ⊗ u2 ( z ) ⊗ u3 ( z ) ⊗ u4 ( z ) = U1,2 ( z ) ⊗ U 3,4 ( z )
min min min min
where
U1,2 ( z ) = u1( z ) ⊗ u2 ( z ) and U 3,4 ( z ) = u3 ( z ) ⊗ u4 ( z ) .
min min
The total capacity of a subsystem containing two independent

components connected in parallel (able to perform the same task) is equal
to the sum of the capacities of the individual components. Therefore, the
structure function for such a subsystem takes the form
G{i, j} = Gi + G j . (3.72)
The u-function representing the pmf of G{i, j} can be, therefore, obtained
using the operator ⊗ over the u-functions representing pmf of Gi and Gj.
+
For example the u-function U1,2(z) representing the total performance of a

subsystem consisting of a pair of two-state components with nominal
performances g1 and g2 and availabilities A1 and A2 connected in parallel

takes the form
U1,2 ( z ) = u1 ( z ) ⊗ u2 ( z ) = ( A1z g1 + A1z 0 ) ⊗( A2 z g 2 + A2 z 0 )
+ +
(3.73)
= A1 A2 z g1 + g 2 + A1 A2 z g 2 + A1 A2 z g1 + A1 A2 z 0 .
As in the case of the series connection, the u-functions for any parallel
system can be obtained recursively by consecutively determining the u-
functions of pairs of the components.
An example of a typical block diagram for a power unit is presented in
Fig. 3.6. In this figure the random capacities of different facilities,
subsystems and an entire power unit are determined in accordance with the
block diagram.
G11
G 31 G 32
G 12 G 22
G12
G 33 … G 1n
… G 23 G 24
G 34 G 25
G1k
k G2 = min(G12 , G22 ) G3 = min(G31 , G32 )

G1 = ∑ G1i Gn = G1n
i =1 + min(G23 , G24 ) + G33 + min(G34 , G35 )
n
G= ∑G j
j =1
Fig. 3.6. Example of power unit block diagram
The complex series-parallel system can always be represented as

composition of the independent subsystems containing only components
connected in a series or in parallel. Therefore, in order to obtain the u-
function of a series-parallel system one has to apply the composition
operators recursively and obtain u-functions of the intermediate pure series
or pure parallel structures.
The following algorithm realizes this approach:
1. Find the subsystem consisting of two components connected in
parallel or in series in the given reliability block diagram.
2. Obtain u-function of these subsystem using the corresponding ⊗ or

min
⊗ operators.
+
3. Replace in the diagram the subsystem with single component having the
u-function obtained for the given subsystem.
4. If the reliability block diagram contains more then one element return to
step
The resulting u-function represents the capacity distribution of the entire
system.
Example 3.5
In order to illustrate the recursive approach (the reliability block diagram
method), consider the series-parallel power station coal feeding system
presented in Fig. 3.7A. The parameters of the system elements are
presented in Table 3.11.
Table 3.11. Parameters of elements of series-parallel system

j 1 2 3 4 5 6 7
Primary First- Second- Third- Bypass Secondary Bypass
feeder level level level stacker feeder conveyor
stacker stacker stacker
gj 5 3 5 4 2 6 3
Aj 0.9 0.8 0.9 0.7 0.6 0.8 0.8
First, one can find a series subsystem consisting of elements with the u-
functions u2(z), u3(z) and u4(z). By calculating the u-functions
U1(z)= u2 ( z ) ⊗ u3 ( z ) and U2(z)= U1( z ) ⊗ u4 ( z ) and replacing the three
min min
elements with a single element with the u-function U2(z) one obtains a
system with the structure presented in Fig. 3.7B. This system contains a
purely parallel subsystem consisting of elements with the u-functions U2(z)
and u5(z), which in their turn can be replaced by a single element with the
u-function U 3 ( z ) = U 2 ( z ) ⊗ u5 ( z ) (Fig. 3.7C). The structure obtained has
+
three elements connected in a series that can be replaced with a single

element having the u-function U 5 ( z ) , where U 4 ( z ) = u1( z ) ⊗ U 3 ( z ) and
min
U 5 ( z ) = U 3 ( z ) ⊗ u6 ( z ) (Fig. 3.7D). The resulting structure contains two

min
elements connected in parallel. The u-function of this structure representing

the pmf of the entire system performance is obtained as

U ( z ) = U 5 ( z ) ⊗ u7 ( z ).
+
u2(z) u3(z) u4(z) U1(z)

u1(z) u6(z) u1(z) u6(z)
u5(z) u5(z)
u7(z) u7(z)
A B
u1(z) U2(z) u6(z) U3(z)
u7(z) u7(z)
C D
Fig. 3.7. Example of recursive determination of the u-function for a series-parallel

system
The process of calculating U(z) for the system is as follows:

3 0 5 0 3 0
⊗ u3(z) =(0.8z +0.2z ) ⊗ (0.9z +0.1z )=0.72z +0.28z
U1(z)=u2(z) min min
3 0 4 0
⊗ u4(z)= (0.72z +0.28z ) ⊗ (0.7z +0.3z )
U2(z)= U1(z) min min
=0.504z3+0.496z0
U3(z)=U2(z) ⊗
+
u5(z)=(0.504z3+0.496z0) ⊗
+
(0.6z3+0.4z0)
= 0.3024z6+0.4992z3+0.1984z0
⊗ U3(z)=(0.9z5+0.1z0) ⊗ (0.3024z6+0.4992z3+0.1984z0)
U4(z)=u1(z) min min
= 0.27216z5+0.44928z3+0.27856z0;
5 3
⊗ u6(z)=(0.27216z +0.44928z
U5(z) = U4(z) min
+0.27856z0) min 6 0 5 3 0
⊗ (0.8z +0.2z )= 0.217728z +0.359424z +0.422848z ;
U(z)=U5(z) ⊗
+
u7(z)=(0.217728z5+0.359424z3+0.422848z0) ⊗
+
(0.8z3
+0.2z0)=0.1741824z8+0.2875392z6+0.0435456z5
+0.4101632z3+0.0845696z0.
Having the system u-function that represents its capacity distribution one
can easily obtain the system mean capacity E(G)=U'(1)=4.567. The system
LOLP for different constant load levels l can be obtained by summing the
probabilities corresponding to system capacity levels that are not less than
the load l in the system capacity pmf represented by the u-function U(z):
LOLP(l) = 0.0846 for 0<l≤3
LOLP(l) = 0.4947 for 3<l≤5
LOLP(l) = 0.5383 for 5<l≤6
LOLP(l) = 0.8258 for 6<l≤8
LOLP(l) = 1 for l>8
3.5 Power plant reliability optimization problems
Reliability improvement is of critical importance in various kinds of

systems; however, any effort for this type of improvement usually requires
resources that are limited by technical and/or economical constraints. Two
approaches can be distinguished in the reliability optimization problem.
The first one is aimed at achieving the greatest possible reliability subject
to different constraints (according to (Gnedenko and Ushakov 1995), this
problem is named the direct reliability optimization problem), and the
second one focuses on minimizing the resources needed for providing a
required reliability level (this problem is named the inverse reliability
optimization problem.
There are four general methods for improving system reliability:
• a provision of redundancy;
• an optimal adjustment of the system’s parameters, an optimal
arrangement of the existing elements or the assignment of
interchangeable elements;
• an enhancement of the reliability (availability) and/or performance of
the system’s elements;
• a combination of the above-mentioned methods.
Applied to power generation systems, these methods affect two basic
system properties: its configuration and the capacity distribution of its
elements.
The UGF method that allows for system capacity distribution, and
thereby its performance measures to be evaluated based on a fast procedure,
opens new possibilities for solving reliability optimization problems. Based
on the UGF technique, the system reliability can be obtained as a function of
the system structure and the capacity distributions of its elements.

Therefore, numerous optimization problems can be formulated in which the
optimal composition of all or part of the factors influencing the entire
system reliability has to be found subject to different constraints.
3.5.1 Structure optimization problems

In binary systems, providing redundancy means to incorporate several iden-
tical parallel elements into a certain functional subsystem (component). The
redundancy optimization problem for a system, which may consist of ele-
ments with different capacity distributions, is a problem of system structure
optimization. Indeed, when solving practical system design problems, the
reliability engineer deals with a variety of products existing on the market.
Each product is characterized by its capacity distribution, price, etc. To find
the optimal system structure, one should choose the appropriate versions
from a list of available products for each type of equipment, as well as the
number of parallel elements of these versions.
In the simplest structure optimization problem, each component can
contain only identical elements. This optimization problem is relevant in
cases where contracting or maintenance considerations prevent purchasing
or using different elements of the same type. In some cases, such a
limitation can be undesirable or even unacceptable for two reasons:
• Allowing different versions of the elements to be allocated in the same
system component, one obtains a solution providing the desired relia-
bility level at a lower cost than what appears in the solution containing
identical parallel elements.
• In practice, when a system needs to be modernized according to new
demand levels or new reliability requirements, the designer often has to
include additional elements in the existing system. Some system compo-
nents can contain versions of elements that are unavailable. In this case,
some elements with the same functionality but with different para-
meters should make up the components. Therefore, in the general case,
the generating system structure optimization problem should be solved
without placing a limitation on the diversity of the versions of the
elements.
The above-mentioned problem of optimal single-stage system expansion
to enhance its reliability and/or performance is an important extension of
the structure optimization problem. In this case, one has to decide which
elements should be added to the existing system and to which component
they should be added. There is a similar problem with the optimal single-
stage replacement of the system elements. Here, one has to decide which
elements should be replaced by new ones having better characteristics.
During the system’s operation time, the demand and reliability require-
ments can change. To provide a desired level of system performance,
management should develop a multistage expansion plan. For the problem
of optimal multistage system expansion, it is important to answer not only
the question of what must be included into the system, but also the question
of when it must be included into the system.
Optimal structure of systems with identical elements in each

component
Problem Formulation. A system consists of N components. Each compo-
nent is a subsystem that can consist of parallel elements with the same
functionality. The interaction between the system components is given by a
reliability block diagram. Different versions of elements may be chosen
for any given system component, but each component can contain only
elements of the same version. For each component i there are Bi element
versions available in the market. A capacity distribution gi(b), pi(b) and cost
ci(b) can be specified for each version b of element of type i.
The structure of system component i is defined by the number of
the element version chosen for this component bi ( 1 ≤ bi ≤ Bi ) and by the
number of parallel elements ni ( 1 ≤ ni ≤ nmax i ), where nmax i is the
maximum allowed number of parallel elements of type i. The vectors b =
(b1, …, bN) and n = (n1, …, nN) define the entire system structure.
For given b and n the total cost of the system can be calculated as
N
C = ∑ ni ci (bi ) (3.74)
i =1
To take into account price discounting, the element cost can be

considered as a function of the number of elements purchased. In this case
N
C = ∑ ni ci (ni , bi ) (3.75)
i =1
The problem of system structure optimization is formulated in (Levitin

et al. 1998) as finding the minimal cost system configuration b, n that
provides the required level A* of the system availability A:
C(b, n)→min subject to f(A,A*) = 1 (3.76)
where f(A,A*) is a function representing the desired relation between A and
A*. If the system should provide a given level of availability, f(A,
A*)=1 ( A ≥ A*).
Implementing the Genetic Algorithm. In order to represent the system

structure in the GA one has to use 2N-length integer strings: a =
(a1,…,ai,…,a2N), where for each i ≤ N bi = aj and for i>N ni−N = ai. An
arbitrary integer string cannot represent a feasible solution because each bi
should vary within limits 1 ≤ bi ≤ Bi and each ni within limits 1 ≤ ni ≤ n max i .
To provide solution feasibility, a decoding procedure should first transform
each string a* to a string a in which
ai = mod Bi (ai* ) + 1 for i ≤ N and ai = mod nmax i (ai* ) + 1 for i > N . (3.77)
The solution decoding procedure, based on the UGF technique, performs

the following steps:
1. Determines ni and bi for each system component from the string a.
2. Determines u-functions uibi (z ) of each version of elements according to
their capacity distribution gi(bi), pi(bi).
3. Determines u-functions of each component i (1≤i≤N) by assigning
U 1 ( z ) = u1b1 ( z ) and applying the recursive equation
U j ( z ) = U j −1 ( z ) ⊗ u ibi ( z ) for j = 2,…,ni (3.78)

φpar
4. Determines the u-function of the entire system U(z) by applying the

corresponding composition operators using the reliability block diagram
method.
5. Having the system u-function, determines its performance measure as
described in the previous chapter.
6. Determines the total system cost using Equation (3.75).
7. Determines the solution’s fitness as a function of the system cost and
performance measure as
M−C(a)−π(1+| A−A*|) (1− f(A,A*)) (3.79)
where π is a penalty coefficient and M is a constant value. The fitness
function is penalized when f(A,A*) = 0. The solution with minimal cost and
with f(A,A*)=1 provides the maximal possible value of the fitness function.
Example 3.6
Consider a power station coal transportation system consisting of five basic
components connected in series (Fig. 3.8):
1. subsystem of primary feeders;
2. subsystem of primary conveyors;
3. subsystem of stackers-reclaimers;
4. subsystem of secondary feeders;
5. subsystem of secondary conveyors.
4 5
1 3
2
Fig. 3.8. Power station coal transportation system
The system belongs to the type of flow transmission system with flow
dispersion, since its main characteristic is the transmission capacity and
parallel elements can transmit the coal simultaneously. The system should
meet a variable demand W. Its acceptability function is defined as
F(G,W)=1( G ≥ W ). The system should have availability not less than
A*=0.99 for the given demand distribution w, q presented in Table 3.12.
Table 3.12. Demand distribution

w 1.00 0.80 0.50 0.20
q 0.48 0.09 0.14 0.29
Each system element is an element with total failure (which means that it
can have only two states: functioning with the nominal capacity or total
failure, corresponding to a capacity of zero). For each type of equipment, a
list of products available on the market exists. Each version of equipment is
characterized by its nominal capacity g, availability p, and cost c. The list
of available products is presented in Table 3.13. The maximal number of
elements in each component should not exceed six.
Table 3.13. Parameters of available system elements

No. of Component 1 Component 2 Component 3
version Primary Primary Stackers-
of system feeders conveyors reclaimers
element g p c g p c g p c
1 1.20 0.980 0.590 1.00 0.995 0.205 1.00 0.971 7.525
2 1.00 0.977 0.535 0.92 0.996 0.189 0.60 0.973 4.720
3 0.85 0.982 0.470 0.53 0.997 0.091 0.40 0.971 3.590
4 0.85 0.978 0.420 0.28 0.997 0.056 0.20 0.976 2.420
5 0.48 0.983 0.400 0.21 0.998 0.042
6 0.31 0.920 0.180
7 0.26 0.984 0.220
Table 3.13. (Cont.)

No. of Component 4 Component 5
version Primary Primary
of system feeders conveyors
element g p c g p c
1 1.15 0.977 0.180 1.28 0.984 0.986
2 1.00 0.978 0.160 1.00 0.983 0.825
3 0.91 0.978 0.150 0.60 0.987 0.490
4 0.72 0.983 0.121 0.51 0.981 0.475
5 0.72 0.981 0.102
6 0.72 0.971 0.096
7 0.55 0.983 0.071
8 0.25 0.982 0.049
9 0.25 0.97 0.044
The total number of components in the problem considered is N = 5.

Each integer string containing 10 integer numbers can represent a possible
solution. In order to illustrate the string decoding process performed by the
GA, consider, for example, the string a* = (9, 10, 4, 1, 3, 0, 7, 6, 1, 2).
From Table 3.13 we have B1 = 7, B2 = 5, B3 = 4, B4 = 9, B5 = 4; nmax i = 6
for any component. After transforming the string according to Equation
(3.77) we obtain a = (3, 1, 1, 2, 4, 1, 2, 1, 2, 3). This string corresponds to
one primary feeder of version 3, two primary conveyors of version 1, one
stacker of version 1, two secondary feeders of version 2, and three
secondary conveyors of version 4.
According to step 2 of the decoding procedure, the u-functions of the
chosen elements are determined as u13(z) = 0.018z0+0.982z0.85 (for the
primary feeder), u21(z) = 0.005z0+0.995z1.00 (for the primary conveyors),
u31(z) = 0.029z0+0.971z1.00 (for the stacker), u42(z) = 0.022z0+0.978z1.00 (for
the secondary feeders) and u54(z) = 0.019z0+0.981z0.51 (for the secondary

conveyor).
According to step 3, we determine the u-functions of the five system
components using the composition operator ⊗ :
+
0 0.85
U1(z) = u13(z) = 0.018z +0.982z
U2(z) = u21(z) ⊗ u21(z) = (0.005z0+0.995z1)2
+
U3(z) = u31(z) = 0.029z0+0.971z1.00

U4(z) = u42(z) ⊗ u42(z) = (0.022z0+0.978z1.00)2
+
U5(z)= u54(z) ⊗ u54(z) ⊗ u54(z) = (0.019z0+0.981z0.51)3

+ +
The availability of flow transmission system consisting of five

components connected in series can be obtained as a product of
availabilities of these components:
4 5
A= ∑ qm ∏ Ai ( wm ),
m =1 i =1
where Ai(wm) is availability of component I for demand wm. The values

Ai(wm) for each system component and each demand level are presented in
Table 3.14.
The overall system availability for the given demand distribution is
A=0.495. The total system cost is
C = 0.47+2×0.205+7.525+2×0.16+3×0.475 = 10.15.
Table 3.14. Availability of system components for different demand levels
w A1(wm) A2(wm) A3(wm) A4(wm) A5(wm)
1.0 0.000000 0.999975 0.971000 0.999516 0.998931
0.8 0.982000 0.999975 0.971000 0.999516 0.998931
0.5 0.982000 0.999975 0.971000 0.999516 0.999993
0.2 0.982000 0.999975 0.971000 0.999516 0.999993
Since the given system configuration provides an availability that is less

than the desired level A*, the solution fitness function is penalized. The
value of the fitness function obtained in accordance with Equation (3.79)
for M=50 and π=25 is 50−10.15−25⋅(1+0.99−0.495) = 2.475.
The best solution obtained by the GA provides the system’s availability
as A = 0.992. According to this solution, the system should consist of two
primary feeders of version 2, two primary conveyors of version 3, three
stackers of version 2, three secondary feeders of version 7, and three
secondary conveyors of version 4. The total system cost is C = 17.05.

Since in this solution f(A, A*) = 1, its fitness is 50−17.05 = 32.95.
Optimal structure of system with different elements in each

component
Problem Formulation. The problem definition is similar to the one
presented in the previous section. However, in this case different versions
and the number of elements may be chosen for any given system
component (Levitin et al. 1997).
The structure of the system component i is defined by the numbers of the
parallel elements of each version b chosen for this component: nib
( 1 ≤ b ≤ Bi ). The vectors ni = (ni1, …, niBi , 1 ≤ i ≤ N ) define the entire
system structure. For a given set of vectors {n1, n2, …, nN} the total cost of
the system can be calculated as
N Bi
C (n1, … , n N )= ∑ ∑ nib ci (b) (3.80)
i=1b=1
Having the system structure defined by its components’ reliability block

diagram and by the set {n1, n2,…, nN}, one can determine the entire system
performance measure O(w, q, n1, n2, …, nN) for any given demand
distribution w, q. The problem of system structure optimization is
formulated as finding the minimal cost system configuration {n1, n2, …,
nN} that provides the required level A* of the system availability A:
C(n1, n2, …, nN) → min subject to f(A,A*) = 1, (3.81)
Implementing the Genetic Algorithm. The natural way of encoding the
solutions of the problem (3.81) in the GA is by defining a B-length integer
string, where B is the total number of versions available:
N
B = ∑ Bi (3.82)
i =1
Each solution is represented by string a = (a1, …, aj, …, aB), where for

each
i −1
j = ∑ Bm + b , (3.83)
m =1
aj denotes the number of parallel elements of type i and version b: nib = aj.
One can see that a is a concatenation of substrings representing the vectors
n1, n2,…, nN.
The solution decoding procedure, based on the UGF technique, performs

the following steps:
1. Determines nib for each system component and each element version
from the string a.
2. Determines u-functions uib ( z ) of each version of elements according
to their capacity distribution gi(b), pi(b).
3. Determines u-functions of subcomponents containing the identical
elements by applying the composition operators ⊗φpar over
nib identical u-functions u ib ( z ).
4. Determines u-functions of each component i (1≤i≤N) by applying the
composition operators ⊗φpar over the u-functions of all nonempty
subcomponents belonging to this component.
5. Determines the u-function of the entire system U(z) by applying the
reliability block diagram method.
6. Determines the system performance measure from the system
performance pmf represented by the u-function U(z).
7. Determines the total system cost using Equation (3.80).
8. Determines the solution fitness as a function of the system cost and
performance measure according to Equation (3.79).
Example 3.7
Consider the coal transportation system described in Example 3.6 and allow
each system component to consist of different versions of the elements.
The total number of available versions in the problem considered is
B = 7+5+4+9+4 = 29. Each string containing 29 integer numbers can
represent a possible solution. In order to illustrate the string decoding
process performed by the GA, consider the string
(0, 0, 0, 2, 0, 0, 1, 0, 0, 2, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 1, 0, 0).
This string corresponds to two primary feeders of version 4, one primary
feeder of version 7, two primary conveyors of version 3, one stacker of
version 1, three secondary feeders of version 7, and one secondary
conveyor of version 2.
According to step 2 of the decoding procedure, the u-functions of the
chosen elements are determined as u14(z) = 0.022z0+0.978z0.85, u17(z) =
0.016z0+0.984z0.26 (for the primary feeders), u23(z) = 0.003z0+0.997z0.53 (for
the primary conveyors), u31(z) = 0.029z0+0.971z1.00 (for the stacker), u47(z)
= 0.017z0+0.983z0.55 (for the secondary feeders) and u52(z) =
0.016z0+0.984z1.28 (for the secondary conveyor).
According to steps 3 and 4, we determine the u-functions of the five

system components using the composition operator ⊗ :
+
U1(z) = u14(z) ⊗ u14(z) ⊗ u17(z)

+ +
0 0.85 2
= (0.022z +0.978z ) (0.016z0+0.984z0.26)
U2(z) = u23(z) ⊗ u23(z) = (0.003z0+0.997z0.53)2
+
U3(z) = u31(z) = 0.029z0+0.971z1.00

U4(z) = u47(z) ⊗ u47(z) ⊗ u47(z) = (0.017z0+0.983z0.55)3
+ +
U5(z) = u52(z) = 0.016z0+0.984z1.28

The entire system availability can be calculated the same way as in the
previous example. We obtain A = 0.95 for the given system structure and
demand distribution.
The total system cost, according to Equation (3.80), is
C = 2×0.42+0.22+2×0.091+7.525+3×0.071+0.825 = 9.805
The fitness of the solution is estimated using Equation (3.79), where M =
50 and π = 25. For the desired value of system availability A* = 0.99 the
fitness takes the value
50−9.805−25(1+0.99−0.95) = 14.195
For the desired value of system availability A* = 0.95, the fitness takes
the value
50−9.805 = 40.195
The minimal cost solutions obtained by the GA for different desired
availability levels A* are presented in Table 3.15. This table presents the
cost, calculated availability, and structure of the minimal cost solutions
obtained by the GA. The structure of each system component i is
represented by a string of the form ni1*bi1, ..., nim*bim, where nim is the
number of identical elements of version bim belonging to this component.
Consider, for example, the best solution obtained for A* = 0.99. The
minimal cost system configuration that provides the system availability A =
0.992 consists of two primary feeders of version 4, one primary feeder of
version 6, two primary conveyors of version 3, two stackers of version 2,
one stacker of version 3, three secondary feeders of version 7, and three
secondary conveyors of version 4. The cost of this configuration is 15.870,
which is 7% less than the cost of the optimal configuration with identical
parallel elements obtained in Example 3.6.
Table 3.15. Optimal solution for system structure optimization problem

A* = 0.95 A* = 0.97 A* = 0.99
System availability A 0.950 0.970 0.992
System cost C 9.805 10.581 15.870
System structure
Primary feeders 1*7, 2*4 2*2 2*4, 1*6
Primary conveyors 2*3 6*5 2*3
Stackers-reclaimers 1*1 1*1 2*2, 1*3
Secondary feeders 3*7 6*9 3*7
Secondary conveyors 1*2 3*3 3*4
3.5.2 Optimal single-stage system expansion
In practice, the designer often has to include additional elements in the

existing system. It may be necessary, for example, to modernize a system
according to new demand levels or new reliability requirements. The
problem of minimal cost system expansion is very similar to the problem of
system structure optimization (Levitin et al. 1997). The only difference is
that each system component already contains some working elements. The
initial structure of the system is defined as follows: each component of type
i contains B'i different subcomponents connected in parallel. Each
subcomponent j in its turn contains n'ij identical elements, which are also
connected in parallel. Each element is characterized by its capacity
distribution g'i(j), p'i(j). The entire initial system structure can, therefore, be
defined by a set {g'i(j), p'i(j), n'ij: 1≤i≤N, 1≤j≤B'i} and by a reliability block
diagram representing the interconnection among the components.
The optimal system expansion problem formulation and the GA
implementation are the same as in the case of system structure optimi-
zation. The only difference is that for the availability evaluation one should
take into account u-functions of both the existing elements and the new
elements chosen from the list, while the cost of the existing elements should
not be taken into account when the system expansion cost is calculated.
Example 3.8
Consider the same coal transportation system that was presented in
Example 3.6. The initial structure of this system is given in Table 3.16.
Each component contains a single subcomponent consisting of identical
elements: B'i = 1 for 1≤i≤5. The existing structure can satisfy the demand
presented in Table 3.12 with the availability A(w,q)=0.506. In order to
increase the system’s availability to the level of A*, the additional elements
should be included. These elements should be chosen from the list of

available products (Table 3.13).
Table 3.16. Parameters of initial system structure

Component Capacity Availability Number of parallel
no. elements
1 0.75 0.988 2
2 0.28 0.997 3
3 0.66 0.972 2
4 0.54 0.983 2
5 0.66 0.981 2
The minimal cost system expansion solutions for different desired values
of system availability A* are presented in Table 3.17.
Table 3.17. Best obtained solutions for system expansion problem
A* = 0.95 A* = 0.97 A* = 0.99

System availability 0.950 0.971 0.990
Expansion cost 0.630 3.244 4.358
Added elements
Primary feeders − 1*6 1*6
Primary conveyors 2*5 1*5, 1*4 1*5
Stackers- − 1*4 1*3
reclaimers
Secondary 1*4 1*4 1*4
conveyors
Consider, for example, the best solution obtained for A* = 0.99, encoded
by the string (0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0,
0, 0, 0, 0, 1). The corresponding minimal cost system expansion plan that
provides the system availability A = 0.990 presumes the addition of the
primary feeder of version 6, the primary conveyor of version 5, the stacker
of version 3, the secondary feeder of version 7 and the secondary conveyor
of version 4. The cost of this expansion plan is 4.358.
3.5.3 Optimal multistage system expansion
In many cases, when the demand and/or reliability requirements increase

with time, the design problem concerns multiple expansions
(reinforcements) rather than the construction of complete new systems or
their single-stage expansions. While in the problems of system structure
optimization or single-stage expansion it is sufficient to ask what should be
done and where, in designing the expansion plan an answer is also needed
to the question of when.
Using the single-stage system expansion algorithm, one can consider the
reinforcement planning problem as a sequence of expansions so that each
expansion is a separate problem. However, this may not lead to the
optimization of the whole plan, because, in general, partial optimal
solutions cannot guarantee an overall optimal solution. Therefore, a more
complex method should be developed to solve the system-expansion
planning problem in several stages, rather than in a single stage (Levitin
2000).
This section addresses the multistage expansion problem for systems.
The study period is divided into several stages. At each stage, the demand
distribution is predicted. The additional elements chosen from the list of
available products may be included into any system component at any
stage to increase the total system availability. The objective is to minimize
the sum of costs of investments over the study period while satisfying
availability constraints at each stage.
Problem Formulation. As in the single-stage expansion problem, the

initial system structure is defined by a reliability block diagram
representing the interconnection among system components and by a set
{g'i(j), p'i(j), n'ij : 1≤i≤N, 1≤j≤B'i}.
In order to provide the desired level of availability, the system should be
expanded at different stages of the study period. Each stage y (1≤y≤Y)
begins τ(y) years after the base stage (stage 0) and is characterized by its
demand distribution w(y), q(y). The timing of stages can be chosen for
practical reasons so that τ(y) must not be evenly spaced. Increasing the
number of stages Y during the planning horizon increases the solution
flexibility. Indeed, the addition of intermediate stages may allow invest-
ments to be postponed without violating the system availability constraints.
On the other hand, the increase of number of stages expands drastically the
search space and, therefore, slows the algorithm convergence. The
recommended strategy is to obtain a number of solutions for different Y in a
reasonable time and to choose the best one.
Different versions and numbers of elements may be chosen for

expansion of any given system component. The versions can be chosen
from the list of elements available in the market. The expansion of system
component i at stage y is defined by the numbers nib ( y ) of parallel
elements of each version b (1≤b≤Bi) included into the component at this
stage The set ny={ nib ( y ) : 1≤i≤N, 1≤b≤Bi} defines the entire system
expansion at stage y.
For a given set ny the total cost of the system expansion at stage y can be
calculated in present values as
1 N Bi
C ( y )= ∑ ∑ nib ( y)ci (b) (3.84)
(1 + IR) τ ( y ) i=1b=1
where IR is the interest rate. The total system expansion cost for the given
expansion plan {n1,…,nY} is
Y 1 N Bi
C= ∑ ∑ ∑ nib ( y)ci (b) (3.85)
y =1 (1 + IR )τ ( y ) i=1b=1
The system structure at stage y (after expansion) is defined by a set of

elements composing the initial structure n' = {n'ij : 1≤i≤N, 1≤j≤B'i} and by
a set of included elements my = {mib ( y ) : 1≤i≤N, 1≤ b ≤Bi}, where
y
mib ( y ) = ∑ nib (k ) (3.86)
k =1
Having the system structure and demand distribution for each stage y,
one can evaluate the system performance measures. The problem of system
expansion optimization can be formulated as follows:
Find the minimal cost of system expansion plan that provides the
required level A* of the system availability Ay at each stage y:
C (n1,…,nY) → min
subject to f(Ay(n1,…,nY, w(y), q(y)), A*) =1 for 1≤ y≤Y (3.87)
where f(Ay,A*) is a function representing the desired relation between Ay
and A*.
The problem of system structure optimization can be considered as a
special case of this formulation. Indeed, if the set n' of the elements
composing the system at stage 0 is empty and Y=1, then the problem is
reduced to system structure optimization subject to reliability constraints. If
n' is empty but Y≠1, then the problem is that of multistage system structure
optimization, which includes both determination of the initial system

structure and its expansion plan.
Implementing the Genetic Algorithm. The natural way to represent the

expansion plan n1,…,nY is to use a string containing integer numbers
corresponding to nib ( y ) for each type i, version bi and stage y. Such a
string contains Y ∑iN=1 Bi elements, which can result in an enormous growth
of the length of the string even for problems with a moderate number of
available versions of elements and number of stages. Besides, to represent a
reasonable solution, such a string should contain a large percentage of
zeros, because only small number of elements should be included into the
system at each stage. This redundancy causes an increase in the need for
computational resources and lowers efficiency of the GA.
In order to reduce the redundancy of expansion plan representation, each
inclusion of n elements of version b into the system at stage y is
represented by the triplet {n, b, y}. The maximum number of possible
inclusions into each component (diversity factor) K and the maximum
number of identical elements in each inclusion nmax are defined as
parameters which can be preliminarily specified for each problem. Thus,
the first K triplets represent the total expansion plan for the first
component, the second K triplets for the second component, and so on. The
length of string representing a solution is in this case 3KN, which can be
much smaller than that in the first case.
Consider a string a = (a1,...,a3KN) where each element aj is generated in
the range
0 ≤ a j ≤ max{n max , Y , max Bi } (3.88)
1≤i≤ N
For each system component i, each of K triplets (ax,ax+1,ax+2), where

x = 3K(i−1)+3(j−1) for 1≤j≤K (3.89)
determines the addition of mod nmax +1 (a x ) elements of version
mod Bi (a x +1 ) + 1 at stage mod Y (a x +2 ) + 1. This transform provides solution
feasibility by mapping each of the variables into a range of its possible
values, since nib ( y ) varies from 0 to nmax, b varies from 1 to Bi and y varies
from 1 to Y.
For example, consider a problem with K = 3, Y = 5, nmax = 4, where
max {Bi } = 9. Substring (7, 3, 2, 0, 2, 6, 1, 7, 0) corresponding to a
1≤ i ≤ N
component for which eight different versions are available, represents

addition of mod4+17 = 2 elements of version mod83+1 = 4 at stage
mod52+1 = 3 and mod4+11 = 1 element of version mod87+1 = 8 at stage

mod50+1 = 1. The second triplet (0, 2, 6) represents addition of zero
elements and, therefore, should be discarded. The existence of such
“dummy” triplets allows the number of additions to each component to
vary from 0 to K, providing flexibility of solution representation.
Having the vector a, one can determine the system structure in each
stage y using (3.86) and obtain the system capacity distribution using the
UGF technique. For this capacity distribution and for the given demand
distribution w(y), q(y) one obtains the system availability Ay. The solution
fitness for the estimated values of Ay and C calculated using (3.85) should
be determined as
Y
M − C (a ) − π ∑ [(1+ | A* − A y (a ) |)(1 − f ( A y (a ), A* )] (3.90)
y =1
Example 3.9
In this example, we consider the same coal transportation system from
Example 3.6. The initial system structure is presented in Table 3.16. Table
3.18 contains the boiler system demand distributions at five different stages
and times from the present to the beginning of these future stages.
Because of demand increase, the availability index of the system
becomes A1(w(1), q(1)) = 0.506 at the first stage. To provide a desired
availability level at all the stages, the system should be expanded. The
characteristics of products available in the market for each type of
equipment are presented in Table 3.13.
Table 3.18. Demand distributions
Stage no. τ ( y) Demand distribution
1 0 w 1.00 0.80 0.50 0.20
q 0.48 0.09 0.14 0.29
2 3 w 1.20 0.80 0.50 -
q 0.43 0.32 0.25 -
3 6 w 1.40 1.20 0.80 0.50
q 0.10 0.39 0.31 0.20
4 9 w 1.40 1.20 0.80 -
q 0.41 0.24 0.35 -
5 12 w 1.60 1.40 1.00 -
q 0.30 0.45 0.25 -
The best expansion plans obtained by the GA for different values of A*

are presented in Tables 3.19 - 3.21. The interest rate for the problems
considered is IR = 0.1. Expansion of each system component i at stage y is

presented in the form n*b, where n is a number of identical elements of
version b to be included into the component at a given stage. The table also
contains system availability Ay obtained at each stage, total expansion costs
in present values C, and costs of the system expansions at the first stage
C(1).
Table 3.19. Best obtained multistage expansion plan for A* = 0.95

Expansion cost C = 4.127, C(1) = 0.645
Stage 1 2 3 4 5
System availability 0.951 0.951 0.963 0.952 0.960
Added elements
Primary feeders 1*4
Primary conveyors 2*5 1*5 1*5
Stacker-reclaimers 1*4 1*2
Secondary feeders 1*7 1*7
Secondary conveyors 1*3 1*4

Stage 1 2 3 4 5
System availability0.972 0.970 0.987 0.975 0.970
Added elements
Primary feeders 1*4 1*7
Primary conveyors 3*5 2*5
Secondary feeders 2*7

Stage 1 2 3 4 5
System availability0.995 0.991 0.990 0.996 0.994
Added elements
Primary feeders 1*6 1*6 2*6
Primary conveyors 1*3 2*4
Secondary feeders 1*7 2*7
One can compare expansion at the first stage of multistage plans with the
single-stage expansion plans presented in Table 3.17. The demand
distribution for the single-stage problem is the same as that in the first stage
of the multistage problem. The solutions for the first stage of the multistage
plan differ from ones that are obtained for single-stage expansion. The
comparison shows that a single-stage expansion plan is less expensive than
the first stage of a multistage plan, which should consider the effect of the
first stage expansion on the further stages.
To demonstrate the advantages offered by incremental expansion, the
solutions that satisfy reliability constraints for the final stage demand
distribution but require all expansion investments at time zero were
obtained (Table 3.22). One can see that incremental expansion provides
considerably less expensive solutions.
Table 3.22. Best obtained single-stage expansion plans

A* = 0.95 A* = 0.97 A* = 0.99
System availability 0.951 0.970 0.990
Expansion cost 8.660 8.895 10.192
Added elements
Primary feeders 1*7 1*4 2*6, 1*7
Primary conveyors 4*5 1*3, 2*4 5*5
Stacker-reclaimers 2*3 2*3 1*2, 1*3
Secondary conveyors 2*4 2*4 2*4
3.5.4 Optimal multistage system modernization
When the system demand and/or reliability requirements increase over

time, multiple system modernization actions are usually performed.
Besides the system’s multistage expansion the system modernization plan
can include modification of the existing equipment and changes in the
maintenance policy. Both types of action lead to changes in the capacity
distribution of the elements already presented in the system. The multistage
modernization plan should consider all these possible alternatives.
Problem Formulation. The initial system structure is defined by a

reliability block diagram representing the interconnection among N system
components and by a set Φ(0) = {g'i(j), p'i(j), n'ij: 1≤i≤N, 1≤j≤B'i}. Each
component i can contain up to Ei different elements connected in parallel.
In order to allow number of elements in each component to vary we
introduce “dummy” elements of version 0.
In order to provide the desired levels of the system performance O*y at
different stages y of the study period, the system should be modernized.
Each stage y (1≤y≤Y) begins τ(y) years after the base stage (stage 0) and is
characterized by its demand distribution w(y), q(y). At each stage, three
different types of action may be undertaken: modification of some of the
system’s elements, addition of new elements, and removal of some of the
system’s elements. In order to manipulate the system structure in a uniform
way, we will formulate all of these actions in terms of element
replacements:
The modification of element j of component i or changing the
procedure of the element maintenance leads to changes in its capacity
distribution. We can consider this action as replacement of the element of
version bij with the element of version b*ij. Parameters of new versions of
the element should be specified for each element modification action.
The addition of a new element into the system can be considered as the
replacement of the element of the corresponding type i and version 0 with a
new element with version b*ij. For each component i there are Bi element
versions available in the market.
The removal of the element of version bij from the system can be
formulated as its replacement with the element of version 0.
Now one can define each system modernization action x by vector
{ix, bix x , b*i x , cx} (3.91)
x
where ix is the number of the component in which the element of version

bi x x should be replaced with the element of version b*i x and cx is the cost
x
of the action. Modernization action is feasible if the system component ix

already contains elements of version bix x . Unfeasible actions should be
replaced with “dummy” ones which correspond to x = 0. Action x = 0 is
associated with cost c0 = 0 and implies that no replacements are performed.
The system modernization plan at stage y can be defined by the vector
xy = (xys, 1≤h≤H) where H is the maximal number of possible actions at a
single stage. The cost of plan xy can be calculated in present values as
1 H
C y= ∑ cx yh (3.92)
(1 + IR)τ ( y ) h=1
where IR is the interest rate.

Each modernization plan xy transforms the set of versions of elements
presented in the system at stage y−1 into the set of versions of elements
presented in the system at stage y: (Φ(y−1), xy) → Φ(y) for 1≤y≤Y, where
Φ(0) is the initial structure of the system considered.
The total system modernization cost for the given modernization plan
x1, …, xY is
Y Y 1 H
C= ∑Cy = ∑ τ ( y) ∑ cx yh (3.93)
y=1 y=1 (1 + IR) h=1
Having the system structure Φ(y) and demand distribution w(y), q(y) for
each stage y, one can evaluate the system performance measures. The
problem of optimal system expansion can be formulated as follows (Levitin
and Lisnianski 1999):
Find the minimal cost system modernization plan that provides the
required level A* of the availability Ay at each stage y:
C (x1,…,xY) → min
subject to
f(Ay(x1,…, xy, w(y), q(y)), A*) =1 for 1≤ y≤Y. (3.94)
where f(Ay,A*) is a function representing the desired relation between Ay
and A*.
Implementing the Genetic Algorithm. The natural way to represent the

multistage modernization plan x1, …, xY is to use a string a containing HY
integer numbers corresponding to the numbers of the actions to be
undertaken during the study period. In the string a, each number am for
H(y−1)+1 ≤ m ≤ yH corresponds to the action number m−H(y−1) undertaken
at stage y. All of the numbers should be generated in the range 0≤am≤L,
where L is the total number of available actions. Action with number 0
means “do nothing”; it has no cost and is used in order to allow the total
number of actions to vary. One can see that the solution representation
string a contains sequentially allocated vectors xy of the modernization plan
at each stage y = 1,…, Y.
By having the vector a, one can determine the system structure in each
stage y by replacing the elements of version bi x x with elements of version
b*i x in accordance with the modernization plan xy. Using the u-functions of
x
the elements composing the system at each stage y after the corresponding
replacements, one obtains the entire system capacity distribution by
applying the UGF technique. For this capacity distribution and for the
given demand distribution w(y), q(y) one obtains the system availability Ay.
The solution fitness for the estimated values of Ay and the total cost C
calculated using (3.93) is determined according to Equation (3.90).
Example 3.10
Consider two different optimization problems applied to the coal
transportation system from Example 3.6. In the first problem we have to
determine the initial system structure as well as its modernization plan. In
this case, the initial structure set is empty: Φ(0) = ∅. In the second problem
we have to find the optimal modernization plan for the system with the
initial structure presented in Table 3.23. Table 3.24 contains the demand
distributions at five different stages and times from the present to the
beginning of these future stages.
Table 3.23. Initial coal transportation system structure

No. of system component 1 2 3 4 5
Versions of elements, present in the component 1, 1 2, 2, 2 3, 3 4, 4 5, 5
The demand distributions have the same maximal load at stages 1-3 and
4-5. Within these periods, variations of the load demand are caused by
intensification of boiler use. At stage 4, the installation of a new boiler is
planned, which should be supplied by the same coal supply system.
Because of increased demand, the availability index of the system
becomes A1(w(1), q(1)) = 0.506 at the first stage. To provide a desired
availability level A* = 0.95 at all of the stages, the system should be
modernized. The characteristics of the available modernization actions are
presented in Table 3.25 in the form (3.91). The action costs cx comprise
both the investment and operational costs.
Table 3.24. Demand distributions

Stage τ ( y) Demand distribution
no.
1 0 w 1.0 0.8 0.5 0.2
q 0.48 0.09 0.14 0.29
2 3 w 1.0 0.8 0.5 -
q 0.63 0.12 0.25 -
3 6 w 1.0 0.9 0.8 -
q 0.65 0.19 0.16 -
4 9 w 2.0 1.6 0.8 -
q 0.41 0.24 0.35 -
5 12 w 2.0 1.7 1.2 0.9
q 0.52 0.15 0.13 0.20
Each action leads to the appearance of an element of a certain version

either purchased in the market or obtained by modification of some existing
element or its maintenance procedure (equipment removal actions are not

considered in this example).
Table 3.25. List of available actions
x ix bix x b*i x
x
cx Description
1 1 0 6 18.0 Addition of new feeder of version 6
3 1 1 8 5.0 Electric drive system modification aimed
at providing feeder speed increase
4 1 1 9 2.0 Installation of engines monitoring system
in feeder of version 1
5 1 6 10 1.6 Installation of engines monitoring system
6 2 0 11 9.1 Addition of new conveyor of version 11
9 2 2 14 0.8 Changes in mechanical equipment to provide
load increase
10 2 14 15 0.3 Installation of engines monitoring system in
conveyor of version 14
11 2 14 16 0.5 Reduction of flat-belt condition inspection
(replacement) period in conveyor of version 14
13 3 0 3 567.0 Addition of new stacker of version 3
14 3 0 18 359.0 Addition of new stacker of version 18
15 3 3 19 6.0 Installation of hydraulics monitoring system in
stacker of version 3
17 4 4 21 2.0 Reduction of engines inspection period
18 4 20 22 1.8 Reduction of engines inspection period
19 5 0 23 47.5 Addition of new conveyor,
version 23
21 5 5 25 1.4 Reduction of flat-belt inspection (replacement)
period
22 5 25 26 1.1 Action 20 performed after action 21
23 5 24 26 1.4 Action 21 performed after action 20
The list of parameters of different versions of elements is presented in

Table 3.26. The table also contains the origins of the versions (P corres-
ponds to versions already present in the system, N corresponds to new
elements available on the market, and M corresponds to versions obtained
by the modification of existing versions or their maintenance procedures).
Table 3.26. Parameters of available system elements
No. of g p Component Origin
version
1 0.75 0.998 1 P
2 0.28 0.997 2 P
3 0.66 0.972 3 P
4 0.54 0.983 4 P
5 0.66 0.981 5 P
6 0.31 0.920 1 N
7 0.85 0.978 1 N
8 0.93 0.970 1 M
9 0.75 0.995 1 M
10 0.31 0.985 1 M
11 0.53 0.997 2 N
12 0.28 0.997 2 N
13 0.21 0.988 2 N
14 0.33 0.980 2 M
15 0.33 0.991 2 M
16 0.33 0.992 3 M
17 0.21 0.999 3 M
18 0.40 0.971 3 N
19 0.66 0.980 3 M
20 0.55 0.983 4 N
21 0.54 0.998 4 M
22 0.55 0.995 4 M
23 0.51 0.981 5 N
24 0.66 0.990 5 M
25 0.66 0.988 5 M
26 0.66 0.997 5 M
The best modernization plans obtained by the suggested algorithm for

the first and the second problems are presented in Tables 3.27 and 3.28
respectively. The tables contain the list of actions to be undertaken for each
stage, the structure of each of the system’s components, and the system’s
availability index (if there are a few identical actions or elements, they are
presented in the form n*x, where n corresponds to the number of identical
actions of type x). For the interest rate IR = 0.1, the total modernization
costs for problems 1 and 2 are in the present values C = 1783.23 and C =
542.55 respectively.
Table 3.27. Optimal modernization plan for problem the with Φ(0) = ∅
Stage List of Structure of system components
No. actions A 1 2 3 4 5
1 1, 2*2, 2*6, 13, 0.958 6, 2*7 2*11 3, 18 3*20 3*23
14, 3*16, 3*19
2 15 0.951 6, 2*7 2*11 18, 19 3*20 3*23
3 2, 6, 18 0.950 6, 3*7 3*11 18, 19 2*20, 3*23

22
4 2*6, 2*13, 0.952 6, 3*7 5*11 2*3, 4*20, 5*23
2*16, 2*19 18, 19 22
5 5, 2*15, 18 0.950 10, 3*7 5*11 18, 3*20, 5*23
3*19 2*22
Table 3.28. Optimal modernization plan for problem the with Φ(0) ≠ ∅
Stage List of Structure of system components
No actions A 1 2 3 4 5
1 6, 16, 0.950 2*1 3*2, 11 2*3 2*4, 20 2*26
2*21, 2*22
2 4, 2*15 0.951 1, 9 3*2, 11 2*19 2*4, 20 2*26
3 2 0.951 1, 7, 9 3*2, 11 2*19 2*4, 20 2*26
4 2, 2*6, 13, 14, 0.955 1, 2*7, 3*2, 3, 18, 2*4, 2*26,

2*16, 3*19 9 3*11 2*19 3*20 3*23
5 2*9, 15 0.951 1, 2*7, 2, 3*11, 18, 2*4, 2*26,
9 2*14 3*19 3*20 3*23
References
Allan R, Billinton R, Abdel-Gawad N (1986) IEEE reliability test system-
extensions to and evaluation of the generating system. IEEE Trans Power Syst,
1(4): 1-7.
Allan R, Billinton R (1992) Power system reliability and its assessment: Part 1.
Background and generating capacity. Power Engineering Journal, 6(4): 191-
196.
Ashworth M (1992) Integrated Demand and Supply Planning System with
Prosceen-II. In Williams, A. F. (Ed.) Energy Engineering, 89(3).
Babb R (1983) POWRSYM Production Costing Program, TVA.
Baleriaux H, Jamoulle E, Guertechin L (1967) Simulation de l’exploitation d’un
de machines thermiques de production d’electricite couple a des station de
pompage. Revue (edition SRBE) 5: 225-245.
Billinton R and Li W (1992) Consideration of multi-state generating unit models in
composite system adequacy assessment using Monte Carlo simulation.
Canadian Journal of Electrical and Computer Engineering 17(1).
Billinton R and Li W (1994) Reliability Assessment of Electric Power System
Using Monte Carlo Methods. Plenum Press, New York.
Billinton R and Wang P (1999) Deregulated power system planning using a
reliability network equivalent technique. IEE Proc. Gener. Transm. Distrib.
146(1): 25-30.
Booth R (1972) Power system simulation model based on probability analyses.
IEEE Transactions on Power Apparatus and System 9: 62-69.
Ding Y, Wang P and Lisnianski A (2006) Optimal reserve management for
restructured power generating systems. Reliability Engineering and System
Safety 91: 792-799.
Fleck W, Charny L (1984) Electric generation expansion analysis system. Final
report. EPRI-EL-2561-6, Stone and Webster Engineering Corp., Boston, MA
(USA)
Gnedenko B, Ushakov I (1995) Probabilistic reliability engineering. John Wiley
and Sons. New York.
Gurevich V, Or O, Frant Sh. (1992) Maintenance scheduling optimization
Program. Internal Report RD-572, Israel Electric Corp.
Gurevich V, Frant Sh, Bianu A (1995). A generation system reliability model with
Monte Carlo simulation. Proceedings of 18th IEEE convention in Israel.
IEEE Task Force. (1979) IEEE reliability test system. IEEE Trans. Power.
Apparatus Syst. PAS-98: 2047-54.
Jenkins R, Joy D (1974) WIEN Automatic system planning package (WASP).
Oak-Ridge National Laboratory.
Levitin G, Lisnianski A, Elmakis D (1997) Structure optimization of power system
with different redundant elements. Electr Pow Syst Res 43: 19-27.
Levitin G, Lisnianski A, Ben-Haim H, Elmakis D (1998) Redundancy optimization
for series-parallel multi-state systems. IEEE Trans Reliab 47: 165-72.
Levitin G, Lisnianski A (1999) Optimal multistage modernization of power system
subject to reliability and capacity requirements. Electr Pow Syst Res 50: 183-
90.
Levitin G (2000) Multi-state series-parallel system expansion scheduling subject to

reliability constraints. IEEE Trans Reliab 49: 71-79.
Lisnianski A, Levitin G (2003) Multi-state system reliability. Assessment,
optimization and applications. World Scientific, Singapore.
Wang P, Billinton R (2003) Reliability assessment of a restructured power system
using reliability network equivalent techniques. IEE Proc. Gener. Transm.
Distrib. 150(5): 555-560.

03 Reliability of Generation Systems

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

03 Reliability of Generation Systems

Uploaded by

Copyright:

Available Formats

3 Reliability of Generation Systems

The generated electric energy, which is transferred and distributed to the

D. Elmakias: Reliability of Generation Systems, Studies in Computational Intelligence (SCI) 111,

system operation at a predetermined risk level, can be studied by means of

working out the cost-saving reserve management during system operation.

3.1 Reliability indices of generation systems

3.1.1 Calculating reliability indices using the loss of load

In order to calculate the reliability indices of the generation system, it is

generation and the load corresponding to the failure states) can be

the u-function representing the pmf of LLF(G,L):

The expected LLF(G,L) or the LOLP can now be obtained as

The expectation of this time is

The Expected Unsupplied Load (EUL) is the expected value of the

The Expected Unsupplied Energy (EUE) is obtained as the product of

Table 3.1. Distributions of the generation capacities and the load

The u-functions representing the capacity distribution of the generating

= 0.03z1 + 0.21z1 + 0.18 z 0 + 0.18 z 0 + 0.02 z1 + 0.14 z 0

LOLP2 = U 'LLF 2 (1) = 0.26

The u-functions corresponding to the pmf of the UL(G,L) function are

= (0.1z 0 + 0.6 z 0.6 + 0.3z1 ) ⊗ (0.6 z 0.5 + 0.4 z 0.3 )

= 0.06 z max(0.5 − 0,0) + 0.36 z max(0.5 − 0.6,0) + 0.18 z max(0.5 −1,0)

= 0.03z 0.5 + 0.21z 0.1 + 0.18 z 0 + 0.18 z 0 + 0.02 z 0.3 + 0.14 z 0

The Expected Unsupplied Load is

due to failures and/or maintenance and inspection actions leading to partial

The u-function representing the capacity distribution of the entire

UG(z)=UG1(z) ⊗ UG2(z) ⊗ UG3(z)=(0.02z0+0.98z50)

=(0.02z +0.98z )(0.02z +0.02z50+0.96z100)(0.03z0+0.01z50+0.02z100

= 10-4 (0.12 z 0 + 6.04 z 50 + 13.72 z100 + 293.88 z150 + 289.84 z 200

= 0.6 ⋅10 -4 (0.12 z1 + 6.04 z1 + 13.72 z1 + 293.88 z1 + 289.84 z1

The u-function corresponding to the pmf of the unsupplied load function

= 10 (0.12 z + 6.04 z 50 + 13.72 z100 + 293.88 z150 + 289.84 z 200

+ 552.88 z 250 + 8843.52 z 300 ) ⊗ (0.6 z 240 + 0.4 z110 )

The Expected Unsupplied Load is

3.2 Evaluating generation system reliability with respect

There is a growing need within electric utility system operation to refine

for calculating the reliability parameters when considering the system

thus suffering from the lack of information about transitional probabilities

and the expected value of unsupplied energy is calculated as :

In the Monte-Carlo method it is presumed that each scenario has the

where LOLP is the estimate of loss of load probability, EUE is the

V( EUE ) = V(EUE) / M, (3.24)

These expressions indicate that the uncertainty of the estimates depends

β LOLP = LOLP / LOLP (3.27)

β EUE = EUE / EUE (3.28)

Rewriting these expressions with respect to M gives

This expression illustrates some advantages and limitations of the

3.2.1 Monte-Carlo sampling of the system availability states

Scenario of the generation system capacity availability consists of N

determine the vector pnτ.

3.2.2 Implementing unit commitment

where j is index of unit, t is time instant, OCjt is available capacity of unit j

τs time from switching-on the unit j if it has been initially off-line

If interval Ik is the first one (k = 1) and it begins at t0 then the following

Table 3.3. Parameters of generating units

Nominal Minimum Min Min Start

Table 3.4. The transution probabilities matrices

Table 3.4. (cont.)

Fig. 3.2. Transition diagram of generating unit

Table 3.5. Availability of generating units

Table 3.6. Dispatch of generating units during 3 hours.

Table 3.7. Spinning reserve contribution of generating units

Unit 4, though available, cannot be started up immediately due to its

3.3 Optimal reserve management for power generation

The purpose of power generating systems is to provide reliable and